Graph Neural Networks for Recommender Systems: Challenges, Methods, and Directions

09/27/2021 ∙ by Chen Gao, et al. ∙ Tsinghua University 0

Recommender system is one of the most important information services on today's Internet. Recently, graph neural networks have become the new state-of-the-art approach of recommender systems. In this survey, we conduct a comprehensive review of the literature in graph neural network-based recommender systems. We first introduce the background and the history of the development of both recommender systems and graph neural networks. For recommender systems, in general, there are four aspects for categorizing existing works: stage, scenario, objective, and application. For graph neural networks, the existing methods consist of two categories, spectral models and spatial ones. We then discuss the motivation of applying graph neural networks into recommender systems, mainly consisting of the high-order connectivity, the structural property of data, and the enhanced supervision signal. We then systematically analyze the challenges in graph construction, embedding propagation/aggregation, model optimization, and computation efficiency. Afterward and primarily, we provide a comprehensive overview of a multitude of existing works of graph neural network-based recommender systems, following the taxonomy above. Finally, we raise discussions on the open problems and promising future directions of this area. We summarize the representative papers along with their codes repositories in https://github.com/tsinghua-fib-lab/GNN-Recommender-Systems.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1. Introduction

Recommender system, is a kind of filtering system of which the goal is to present personalized information to users, which improves the user experience and promotes business profit. As one of the typical applications of machine learning driven by the real world, it is an extremely hot topic in both industrial and academia nowadays.

To recap the history of recommender systems, it can be generally divided into three stages, shallow models (Rendle et al., 2009; Koren et al., 2009; Rendle, 2010), neural models (He et al., 2017; Guo et al., 2017; Cheng et al., 2016), and GNN-based models (Ying et al., 2018; Wang et al., 2019b; He et al., 2020). The earliest recommendation models capture the collaborative filtering (CF) effect by directly calculating the similarity of interactions. Then model-based CF methods, such as matrix factorization (MF) (Koren et al., 2009) or factorization machine(Rendle, 2010), were proposed to approach recommendation as a representation learning problem. However, these methods are faced with critical challenges such as complex user behaviors or data input. To address it, neural network-based models (He et al., 2017; Guo et al., 2017; Cheng et al., 2016)

are proposed. For example, neural collaborative filtering (NCF) was developed to extend the inner product in MF with multi-layer perceptrons (MLP) to improve its capacity. Similarly, deep factorization machine (DeepFM) 

(Guo et al., 2017) combined the shallow model factorization machine (FM) (Rendle, 2010) with MLP. However, these methods are still highly limited since their paradigms of prediction and training ignore the high-order structural information in observed data. For example, the optimization goal of NCF is to predict user-item interaction, and the training samples include observed positive user-item interactions and unobserved negative user-item interactions. It means that during the parameter updating for a specific user, only the items interacted by him/her are involved.

Recently, the advances of graph neural networks provide a strong fundamental and opportunity to address the above issues in recommender systems. Specifically, graph neural networks adopt embedding propagation to aggregate neighborhood embedding iteratively. By stacking the propagation layers, each node can access high-order neighbors’ information, rather than only the first-order neighbors’ as the traditional methods do. With its advantages to handle the structural data and to explore structural information, GNN-based methods have become the new state-of-the-art approaches in recommender systems.

To well apply graph neural networks into recommender systems, there are some critical challenges required to be addressed. First, the data input of recommender system should be carefully and properly constructed to graph, with nodes representing elements and edges representing the relations. Second, for the specific task, the component in the graph neural network should be adaptively designed, including how to propagate and aggregate, in which existing works have explored various choices with different advantages and disadvantages. Third, the optimization of the GNN-based model, including the optimization goal, loss function, data sampling, etc., should be consistent with the task requirement. Last, since recommender systems have strict limitations on the computation cost, and also due to GNNs’ embedding propagation operations introduce a number of computations, the efficient deployment of graph neural networks in recommender systems is another critical challenge.

In this paper, we aim to provide a systematic and comprehensive review of the research effort, especially on how they improve recommendation with graph neural networks and address the corresponding challenges. To fulfill a clear understanding, we categorize researches of recommender systems from four perspectives, stage, scenario, objectives, and applications. We summarize the representative papers along with their codes repositories in https://github.com/tsinghua-fib-lab/GNN-Recommender-Systems.

It is worth mentioning that there is one existing survey (Wu et al., 2021b) of graph neural network-based recommender system. However, it is limited due to the following reasons. First, it does not provide extensive taxonomy of recommender systems. Specifically, it roughly divides recommender systems into non-sequential recommendation and sequential recommendation, however, which is not so reasonable. In fact, the sequential recommendation is only one specific recommendation scenario with a special setting of input and output, as pointed out by this survey. Second, it does not provide adequate explanations of the motivations and reasons that the existing works leverage graph neural networks for recommender systems. In this survey, by contrast, we provide a comprehensive understanding of why GNN can be and should be used in recommender system, which can help readers to understand the position and value of this new research field. Third, it does not explain the critical challenges of applying graph neural networks to recommendation and how to address them, which are fully discussed in this survey. Last, since this area is increasingly popular, our survey also introduces a lot of recently published papers not covered by (Wu et al., 2021b).

The structure of this survey is organized as follows. We first introduce the background of recommender systems, from four kinds of perspectives (stage, scenario, objective, application), and the background of graph neural networks, in Section 2. We then discuss the challenges of applying graph neural networks to recommender systems from four aspects, in Section 3. Then we elaborate on the representative methods of graph nerual network-based recommendation in Section 4 by following the taxonomy in the above section. We discuss the most critical open problems in this area and provide ideas of the future directions in Section 5 and conclude this survey in Section 6.

2. Background

2.1. Recommender Systems

Figure 1. An illustration of typical recommender systems (stages, scenarios, objectives, and applications)

2.1.1. Overview

In this section, we present the background of recommender systems from four perspectives: stages, scenarios, objectives, and applications. Specifically, in industrial applications, due to the real-world requirements on system engineering, the recommender systems are always split into three stages, matching, ranking, and re-ranking, forming a standard pipeline. Each stage has different characteristics on data input, output, model design, etc. Besides the standard stages, there are many specific recommendation scenarios with a special definition. For example, in the last twenty years, social recommendation has been attracting attention, defined as improving recommender system based on social relations. Last, different recommender systems have different objectives, of which accuracy is always the most important one as it directly determines the system’s utility. Recently, recommender systems have been assigned other requirements such as recommending diversified items to avoid boring user experience, making sure the system treats all users fairly, protecting user privacy from attack, etc. As for the applications, GNN models can be widely deployed in e-commerce recommendation, point-of-interest recommendation, news recommendation, movie recommendation, music recommendation, etc.

2.1.2. Stages

Figure 2. The typical pipeline of recommender systems.

The item pool, i.e. all the items available for recommender systems, is usually large and can include millions of items. Thus, common recommender systems follow a multi-stage architecture, filtering items stage by stage from the large-scale item pool to the final recommendations exposed to users, tens of items. (Covington et al., 2016; Wilhelm et al., 2018). Generally, a modern recommender system is composed of the following three stages.

  • [leftmargin=*]

  • Matching. This first stage generates hundreds of candidate items from the extremely large item pool (million-level or even billion-level), which significantly reduces the scale. Considering the large scale of data input in this stage, and due to the strict latency restrictions of online serving, complicated algorithms cannot be adopted, such as very deep neural networks (Covington et al., 2016; Kang and McAuley, 2019). In other words, models in this stage are usually concise. That is, the core task of this stage is to retrieve potentially relevant items with high efficiency and attain coarse-grained modeling of user interests. It is worth noting that a recommender system in the real world usually contains multiple matching channels with multiple models, such as embedding matching, geographical matching, popularity matching, social matching, etc.

  • Ranking. After the matching stage, multiple sources of candidate items from different channels are merged into one list and then scored by a single ranking model. Specifically, the ranking model ranks these items according to the scores, and the top dozens of items are selected. Since the amount of input items in this stage is relatively small, the system can afford much more complicated algorithms to achieve higher recommendation accuracy (Kang and McAuley, 2018; Song et al., 2019a; Lian et al., 2018). For example, rich features including user profiles and item attributes can be taken into consideration, and advanced techniques such as self-attention (Kang and McAuley, 2018) can be utilized. Since a lot of features are involved, the key challenge in this stage is to design appropriate models for capturing complicated feature interactions.

  • Re-ranking. Although the obtained item list after the ranking stage is optimized with respect to relevance, it may not meet other important requirements, such as freshness, diversity, fairness, etc. (Pei et al., 2019) Therefore, a re-ranking stage is necessary, which usually removes certain items or changes the order of the list in order to fulfill additional criteria and also satisfy business needs. The main concern in this stage is to consider multiple relationships among the top-scored items (Ai et al., 2018; Zhuang et al., 2018). For example, similar or substitutable items can lead to information redundancy when they are displayed closely in the recommendations.

Fig. 2 illustrates the typical pipeline of a recommender system, as well as the comparisons of the three stages above.

2.1.3. Scenarios

In the following, we will elaborate on the different scenarios of recommender systems, including social recommendation, sequential recommendation, session recommendation, bundle recommendation, cross-domain recommendation, and multi-behavior recommendation.

  • [leftmargin=*]

  • Social Recommendation.

    Figure 3. An illustration of social recommendation. User interactions are affected by both their own preferences and the social factor.

    In the past few years, social platforms have dramatically changed users’ daily life. With the ability to interact with other users, individual behaviors are driven by both personal and social factors. Specifically, users’ behavior may be influenced by what their friends might do or think, which is known as social influence (Cialdini and Goldstein, 2004). For example, users in WeChat***http://wechat.com/ Video platform may like some videos only because of their WeChat friends’ like behaviors. At the same time, social homophily is another popular phenomenon in many social platforms, i.e., people tend to build social relations with others who have similar preferences with them (McPherson et al., 2001). Taking social e-commerce as an example, users from a common family possibly share similar product preferences, such as food, clothes, daily necessities, and so on. Hence, social relations are often integrated into recommender systems to enhance the final performance, which is called social recommendation. Fig. 3 illustrates the data input of social recommendation, of which user interactions are determined by both his/her own preferences and social factors (social influence and social homophily).

  • Sequential Recommendation.

    Figure 4. An illustration of sequential recommendation. Given a user’s historical sequence, recommender system aims to predict the next item.

    In recommender systems, users will produce a large number of interaction behaviors over time. The sequential recommendation method extracts information from these behavioral sequences and predicts the user’s next interaction item, as shown in Fig. 4. To symbolize this problem, for the sequence of items that the user has interacted with in order, the goal of the system is to predict the next item that the user will interact with. In recommender systems, the user’s historical behaviors play an important role in modeling the user’s interest. Many commonly used recommendation methods like collaborative filtering (He et al., 2017) train the model by taking each user behavior as a sample. They directly model the user’s preference on a single item, but sequential recommendation uses the user’s historical behavior sequence to learn timestamp-aware sequential patterns to recommend the next item that the user may be interested in. In sequential recommendation, there are two main challenges. First of all, for each sample, i.e., each sequence, the user’s interest needs to be extracted from the sequence to predict the next item. Especially when the sequence length increases, it is very challenging to simultaneously model the short-term, long-term, and dynamic interests of users. Secondly, in addition to modeling within a sequence, since items may occur in multiple sequences or users have multiple sequences, collaborative signals between different sequences require to be captured for better representation learning.

  • Session-based Recommendation.

    Figure 5. An illustration of session-based recommendation. Given a anonymous short session, recommender system aims to predict the next item.

    In many real-world scenarios, such as some small retailers and mobile stream media (e.g. YouTube and Tiktok), it is impossible or not necessary to track the user-id’s behaviors over a long period of time due to limited storage resources. In other words, user profiles and long-term historical interactions are unavailable, and only the short session data from anonymous users are provided. Hence, conventional recommendation methods (e.g., collaborative filtering) may perform poorly in this scenarios. This motivates the problem of session-based recommendation (SBR), which aims at predicting the next item with a given anonymous behavioral session data as shown in Fig. 5. Distinct from the sequential recommendation, the subsequent sessions of the same user are handled independently in SBR, since the behavior of users in each session only shows session-based traits (Hidasi et al., 2015).

  • Bundle Recommendation.

    Figure 6. An illustration of bundle recommendation.

    The existing recommender systems mainly focus on recommending independent items to users. The bundle is a collection of items, which is an important marketing strategy for product promotion. Bundle recommendation aims to recommend a combination of items for users to consume (Liu et al., 2014; Cao et al., 2017; Pathak et al., 2017; Chen et al., 2019a). Bundle recommendation is very common nowadays in online platforms, e.g., the music playlists on Spotify, the pinboards on Pinterest, the computer sets on Amazon, and the furniture suites on IKEA. It is worth mentioning that bundle recommendations are also used to solve interesting and meaningful problems such as fashion outfits (Li et al., 2020c) and drug packages (Zheng et al., 2021c).

  • Cross-Domain Recommendation.

    Figure 7. An illustration of cross-domain recommendation.

    As an increasing number of users interact with multi-modal information across multiple domains, cross-domain recommendation (CDR) has been proven to be a promising method to alleviate cold start and data sparsity problems (Fu et al., 2019; Gao et al., 2019a; Hu et al., 2018; Kang et al., 2019; Man et al., 2017; Mirbakhsh and Ling, 2015; Zhao et al., 2019a). CDR methods can be roughly divided into two categories, single-target CDR (STCDR) and dual-target CDR (DTCDR) (Cui et al., 2020). CDR methods transfer the information from the source domain to the target domain in one direction; DTCDR emphasizes the mutual utilization of information from both source domain and target domain, which can be extended to multiple-target CDR (MTCDR). Since utilizing information from multiple domains can improve performance, cross-domain recommendation has become an important scenario in recommender systems.

  • Multi-behavior Recommendation.

    Figure 8. An illustration of multi-behavior recommendation.

    Users interact with recommender systems under multiple types of behaviors instead of only one type of behavior. For example, when the user clicks on the video, he/she may also perform behaviors such as collecting or comment. In an e-commerce website, users often click, add into shopping carts, share, or collect a product before purchasing it, as shown in Fig. 8. Although the ultimate goal of the recommender system is to recommend products that users will purchase, the purchase behavior is very sparse compared to the user’s click, sharing, and other behaviors. To symbolize this problem, For each user and item , suppose there are different types of behaviors . For the -th behavior, if the user has a observed behavior, then , otherwise . The goal of the recommender system is to improve the prediction accuracy of a certain type of target behavior .

    For multi-behavior recommendation, generally speaking, there are two main challenges. Firstly, different behaviors have different influences on the target behavior. Some behaviors may be strong signals, and some may be weak signals (Xia et al., 2021c; Xia et al., 2021b). At the same time, this influence is different for each user. It’s challenging to model the influence of these different behaviors on the target behavior accurately. Secondly, it is challenging to learn comprehensive representations from different types of behaviors for items. Different behaviors reflect users’ different preferences for items; in other words, different behaviors have different meanings. In order to obtain a better representation, the meaning of different behaviors needs to be integrated into the representation learning.

2.1.4. Objectives

The most important objective of recommender systems, of course, is accuracy. In the following, we elaborate on the three other important objectives, including diversity, explainability, and fairness.

  • [leftmargin=*]

  • Diversity.

    Figure 9. Illustration of individual-level diversity and system-level diversity.

    Two types of diversity are usually considered in recommender systems, namely individual-level diversity and system-level diversity. Specifically, individual-level diversity is an important objective which measures the dissimilarity of the recommended items for each user since repeated similar items make users reluctant to explore with the system. In other words, individual-level diversity reflects how many topics the recommendation list covers and how balanced is the recommended items distributed on different topics. Here topics depend on the recommendation task, e.g. topics can be different product categories for e-commerce recommendation and different genres for music recommendation (Zheng et al., 2021a). Regarding system-level diversity, it compares the recommendation results of different users and expects them to be dis-similar with each other. In other words, low system-level diversity means always recommending popular items to all the users while ignoring long-tail items. Therefore, system-level diversity is sometimes called long-tail recommendation. Fig. 9 briefly illustrates the two types of diversity and their differences. For both individual-level and system-level diversity, there are two main challenges. First, the signal strength of different items varies greatly. For each user, there exist dominant topics and disadvantaged topics, e.g. a user’s interaction records with electronics might be much more frequent than with clothes. Similarly, the signal strength of long-tail items is also far weaker than popular items. Therefore, it is challenging to recommend relevant content with such weak supervision from either disadvantaged topics or long-tail items for individual-level and system-level diversity, respectively. Second, diversity may sometimes contradict recommendation accuracy, resulting in the accuracy-diversity dilemma; thus it is challenging to balance between the two aspects.

  • Explainability.

    As current recommender systems mostly adopt a deep learning paradigm, they can arouse urgent needs on the explainability of recommendation 

    (Zhang and Chen, 2020). The focus of explainable recommender systems is not only to produce accurate recommendation results but to generate persuasive explanations for how and why the item is recommended to a specific user (Zhang and Chen, 2020). Increasing the explainability of recommender systems can enhance users’ perceived transparency (Sinha and Swearingen, 2002), persuasiveness (Tintarev and Masthoff, 2007) and trustworthiness (Kunkel et al., 2019), and facilitate practitioners to debug and refine the system (Zhang and Chen, 2020). This survey mainly focuses on improving explainability with machine learning techniques. Specifically, past research adopts two different approaches (Zhang and Chen, 2020): one makes efforts to design intrinsic explainable models, ensuring the explainability of recommendation results by designing models with transparent logic (rather than merely “black box”), e.g., Explicit Factor Models (Zhang et al., 2014), Hidden Factor and Topic Model (McAuley and Leskovec, 2013), and TriRank (He et al., 2015). The others compromise slightly: they design post-hoc separate models to explain the results generated by “black box” recommender systems, e.g., Explanation Mining (Peake and Wang, 2018)

    . Currently, there are two challenges. First, representing explainable information requires graph-structural item attributes, which are difficult to model without the power of GNN. Second, reasoning recommendations depend on external knowledge in the knowledge graph, which also poses challenges on the task.

  • Fairness. As a typical data-driven system, recommender systems could be biased by data and the algorithm, arousing increasing concerns on the fairness (Mehrotra et al., 2018; Abdollahpouri et al., 2020; Li et al., 2021c). Specifically, according to the involved stakeholders, fairness in recommender systems can be divided into two categories (Mehrotra et al., 2018; Abdollahpouri et al., 2020; Li et al., 2021c): user fairness, which attempts to ensure no algorithmic bias among specific users or demographic groups (Leonhardt et al., 2018; Li et al., 2021a; Beutel et al., 2019), and item fairness, which indicates fair exposures of different items, or no popularity bias among different items  (Mehrotra et al., 2018; Li et al., 2021c; Abdollahpouri et al., 2017, 2019). Here, we focus on the user fairness, and leave item fairness in the section of diversity for their close connection in terms of interpretations and solutions (Mansoury et al., 2020; Abdollahpouri et al., 2017). Specifically, researchers adopt two methods to enhance fairness: One is directly debiasing recommendation results in the training process (Beutel et al., 2019; Zhu et al., 2018), while the other endeavors to rank items to alleviate the unfairness in a post-processing method (Singh and Joachims, 2018; Geyik et al., 2019). Indeed, increasing evidence shows that the utilization of graph data, e.g., user-user, could intensify concerns on fairness (Dai and Wang, 2021; Rahman et al., 2019; Wu et al., 2021a). Thus, it is challenging to debias unfairness in recommendation in the context of rich graph data. Furthermore, it is even harder to boost user fairness in recommendation from the perspective of graph.

2.1.5. Applications

Recommender systems widely existing in today’s information services, with various kinds of applications, of which the representative ones are as follows.

Product recommendation, also known as E-commerce recommendation, is one the most famous applications of recommender systems. For recommendation models in the e-commerce scenario, the business value is highly concerned. Therefore, it is crucial to handle multiple types of behaviors closely relevant to the platform profit, including adding to cart or purchase. Some works (Ma et al., 2018) propose to optimize the click-through rate and conversion rate at the same time. In addition, in e-commerce platform, products may have rich attributes, such as price (Zheng et al., 2020a), category (Zhou et al., 2018), etc., based on which heterogeneous graphs can be constructed (Luo et al., 2020b). Representative benchmark datasets for product recommendation includes Amazon http://jmcauley.ucsd.edu/data/amazon, Tmall https://tianchi.aliyun.com/dataset/dataDetail?dataId=649.

POI (Point-of-interest) recommendation, is also a popular application that aims to recommend new locations/point-of-interests for users’ next visitation. In Point-of-interest recommendation, there are two important factors, spatial factor and temporal factor. The spatial factor refers to naturally-existed geographical attributes of POIs, i.e., the geographic location. In addition, the users’ visitations are also largely limited by their geographical activity areas since the user cannot visit POIs as easily as browsing/purchasing products on e-commerce websites. Besides, the temporal factor is also of great importance since users’ visitation/check-in behaviors always form a sequence. This motivates the problem of next-POI or successive POI recommendation(Lim et al., 2020; Xie et al., 2016; Feng et al., 2015). Representative benchmark datasets for news recommendation includes Yelp §§§https://www.yelp.com/dataset, Gowalla https://snap.stanford.edu/data/loc-gowalla.html, etc.

News recommendation helps users find preferred news, which is also another typical application. Different from other recommendation applications, news recommendation requires proper modeling of the texts of news. Therefore, methods of natural language processing can be combined with recommendation models for extracting news features better 

(Okura et al., 2017). Besides, users are always interested in up-to-date news and may refuse out-of-date ones. Thus, it is also critical but challenging to fast and accurately filter news from the fast-changing candidate pools. As for news recommendation, MIND dataset (Wu et al., 2020d) is the recently-released representative benchmark dataset.

Movie recommendation is one of the earliest recommender systems. Precisely, Netflix’s movie recommendation competition (Bell and Koren, 2007) motivated many pioneer recommendation researches (Bell and Koren, 2007; Koren, 2009)

. The earlier setting of movie recommendation is to estimate the users’ rating scores on movies, from one to five, which is named

explicit feedback. Recently, the binary implicit feedback has become more popular setting (He et al., 2017; Rendle et al., 2009).

There are also other types of recommendation applications, such as video recommendation (Davidson et al., 2010; Liu et al., 2020e), music recommendation (Van Den Oord et al., 2013; Wang and Wang, 2014), job recommendation (Paparrizos et al., 2011), food recommendation (Min et al., 2019), etc.

2.2. Graph Neural Networks

/ The set of graph nodes/edges
/ The set of users/items
The neighborhood set of graph node
The embedding of graph node in the -th propagation layer
The adjacency matrix of graph
Learnable transformation matrix in the -th propagation layer

Nonlinear activation function

Concatenation operation
Hadamard product
Table 1. Frequently used notations.

With the rapid emergence of vast volumes of graph data, such as social networks, molecular structures, and knowledge graphs, a wave of graph neural network (GNN) studies has sprung up in recent years (Kipf and Welling, 2017; Veličković et al., 2018; Berg et al., 2018; Zhang et al., 2019b; Fout, 2017; Feng et al., 2019)

. The rise of GNN mainly originates from the advancement of convolutional neural network (CNN) and graph representation learning (GRL) 

(Zhou et al., 2020a; Wu et al., 2020c). When applied to regular Euclidean data such as images or texts, CNN is extremely effective in extracting localized features. However, for non-Euclidean data like graphs, CNN requires generalization to handle the situations where operation objects (e.g.

, pixels in images or nodes on graphs) are non-fixed in size. In terms of GRL, it aims to generate low-dimensional vectors for graph nodes, edges, or subgraphs, which represent complex connection structures of graphs. For example, a pioneering work DeepWalk 

(Perozzi et al., 2014) learns node representations by using SkipGram (Mikolov et al., 2013) on a generated path with random walks on graphs. Combining CNN and GRL, various GNNs are developed to distill structural information and learn high-level representations. In later parts, we will introduce several general and primary stages for designing a GNN model to accomplish tasks on graphs, illustrated in Fig. 10. Specifically, section 2.2.1, 2.2.2 and 2.2.3 elaborate how to construct graphs, design specialized and effective graph neural networks and optimize models, respectively.

Figure 10. The overall procedure of implementing a GNN model: construct graphs from data (e.g., table or text), design tailored GNN for generating representations, map representations to prediction results, and further define loss function with labels for optimization.

2.2.1. Graph Construction

We first utilize the unified formula to define a graph, i.e., , where and denote the set of graph nodes and edges respectively, and each edge in joins any number of nodes in . In recent years, GNN-based models focus on designing specialized networks for the following three categories of graphs

  • Homogeneous graph, of which each edge connects only two nodes, and there is only one type of nodes and edges.

  • Heterogeneous graph, of which each edge connects only two nodes, and there are multiple types of nodes or edges.

  • Hypergraph, of which each edge joins more than two nodes.

In many information services nowadays, relational data is naturally represented in the form of graphs. For example, implicit social media relationships can be considered as a unified graph, with nodes representing individuals and edges connecting people who follow each other. However, since non-structured data such as images and texts do not explicitly contain graphs, it is necessary to define nodes and edges manually for building graphs. Taking text data used in Natural Language Processsing (NLP) as an example, words/documents are described as nodes, and edges among them are constructed according to Term Frequency-Inverse Document Frequency (IF-ITF) (Yao et al., 2019). Additionally, an emerging research direction in representation learning on graphs is knowledge graph (KG), which is a representative instance of the heterogeneous graph. KG integrates multiple data attributes and relationships, in which nodes and edges are redefined as entities and relations, respectively. Specifically, the entities in KG can cover a wide range of elements, including persons, movies, books, and so on. The relations are utilized to describe how entities associate with each other. For example, a movie can relate to persons (e.g., actors or directors), countries, languages, etc. Except for ordinary graphs, the hypergraph is also explored recently to handle more complex data (e.g., beyond pairwise relations and multiple modals) flexibly (Feng et al., 2019), in which each edge can connect more than two nodes.

In summary, constructing graphs necessitates either pre-existing graph data or abstracting the concept of graph nodes and edges from non-structured data.

2.2.2. Network Design

Generally speaking, GNN models can be categorized into spectral and spatial models. Spectral models consider graphs as signals and process them with graph convolution in the spectral domain. Specifically, graph signals are first transformed into spectral domain by Fourier transform defined on graphs, then a filter is applied, at last the processed signals are transformed back to spatial domain 

(Shuman et al., 2013). The formulation of processing the graph signal with the filter is

(1)

where denotes the graph Fourier transform.

In contrast, spatial models conduct the convolution on graph structures directly to extract localized features via weighted aggregation like CNNs. Despite the fact that these two types of models start from different places, they fall into the same principle of collecting neighborhood information iteratively to capture high-order correlations among graph nodes and edges. Here “information” is represented as embeddings, i.e., low-dimensional vectors. To this end, the primary and pivotal operation of GNN is to propagate embeddings on graphs following structural connections, including aggregating neighborhood embeddings and fusing them with the target (a node or an edge) embedding, to update graph embeddings layer by layer.

In the following, we will introduce several groundbreaking GNN models to elaborate how neural networks are implemented on graphs. The frequently used notations are explained in Table 1.

  • [leftmargin=*]

  • GCN (Kipf and Welling, 2017). This is a typical spectral model that combines graph convolution and neural networks to achieve the graph task of semi-supervised classification. In detail, GCN approximates the filter in convolution by the first order following (Hammond et al., 2011). Then the node embeddings are updated as follows,

    (2)

    of which the derivation can refer to (Wu et al., 2020c). is the embedding matrix of graph nodes in the -th layer of convolution, where is the embedding dimension. Besides, is the adjacency matrix of the graph with self-loop, of which each entry if the node connects with or ; otherwise , and .

  • GraphSAGE (Hamilton et al., 2017). This is a pioneered spatial GNN model that samples neighbors of the target node, aggregates their embeddings, and merges with the target embedding to update.

    (3)

    where denotes the sampled neighbors of the target node . The function AGGREGATE has various options, such as MEAN, LSTM (Hochreiter and Schmidhuber, 1997) and so on.

  • GAT (Veličković et al., 2018). This is a spatial GNN model that addresses several key challenges of spectral models, such as poor ability of generalization from a specific graph structure to another and sophisticated computation of matrix inverse. GAT utilizes attention mechanisms to aggregate neighborhood features (embeddings) by specifying different weights to different nodes. Specifically, the propagation is formulated as follows,

    (4)

    where is the propagation weight from node to node and is the neighborhood set of node , including itself. As shown in the second equation, the attention mechanism is implemented via a fully-connected layer parameterized by a learnable vector , followed by the softmax function.

  • HetGNN (Zhang et al., 2019b). This is a spatial GNN tailored for heterogeneous graphs. Considering the heterogeneous graph consists of multiple types of nodes and edges, HetGNN first divides neighbors into subsets according to their types. Hereafter, an aggregator function is conducted for each type of neighbor to gather localized information, combining LSTM and MEAN operations. Furthermore, different types of neighborhood information are aggregated based on the attention mechanism. Detailed formulas are omitted since the implementation is following the above works.

  • HGNN (Feng et al., 2019). This is a spectral model implementing GNN on the hypergraph. The convolution is defined as follows,

    (5)

    where each entry of denotes whether the hyperedge contains the node , each diagonal entry of denotes how many hyperedges the node is included in and each diagonal entry of denotes how many nodes the hyperedge includes. Generally, this convolution operation can be considered as two stages of propagating neighborhood embeddings: 1) propagation from nodes to the hyperedge connecting them, and 2) propagation from hyperedges to the node they meet.

The commonality and difference among typical GNN models above is illustrated in Fig. 11.

In order to further capture high-order structural information on the graph, the convolution or embedding propagation mentioned above will be performed for times. In most cases, since GNNs are suffered from the over-smoothing problem that the updated embeddings will be in small fluctuations when the number of propagation layers becomes larger. Relevant studies focusing on developing deep and effective GNN models will be introduced in section 5.1.

Figure 11. Comparison of several typical GNN models. For graph type, nodes and edges type are represented by colors and line-styles respectively. For aggregation, line-width indicates neighborhood weight.

2.2.3. Model Optimization

After the processing of the designed network in section 2.2.2, overall embeddings of nodes or edges encoding feature semantics as well as graph structures are produced. To perform the downstream graph learning tasks, these embeddings will be further transformed to targets (e.g.

, the probability that a node belongs to a class) by general neural networks (

e.g., MLP).

There are mainly classification, prediction and regression tasks on graphs, including three levels: node, edge, and subgraph. Despite the disparity of various tasks, there is a standard procedure for model optimization. Specifically, relevant embeddings will be mapped and come with labels to formulate the loss function, and then existing optimizers are utilized for model learning. Following this process, there are several types of mapping functions (e.g., MLP, inner product) and loss functions (e.g., pair-wise, point-wise) to choose for specific tasks. For pair-wise loss function, the discrimination between positive and negative samples is encouraged, and a typical formulation BPR (Rendle et al., 2009) is

(6)

where

is the sigmoid function.

and denote positive and negative samples respectively, and is for measuring the samples. For point-wise loss function, it includes mean square error (MSE) loss, cross-entropy loss and so on.

For a better understanding, we take link prediction and node classification tasks as examples to elaborate on how the GNN model is optimized. For link prediction, the likelihood that whether an edge exists between two nodes , requires definition. Technically, it is usually calculated based on similarity with node embeddings in each layer of propagation:

(7)

where denotes the mapping function. Furthermore, we can construct the training data as , consisting of observed positive and randomly-selected negative samples, and , respectively. Specifically, the node connects with on the graph but not with . In recommender system, the samples will indicate that user has interacted with item but not interacted with item , where is sampled from all the other items has not interacted before. Hereafter, if the pair-wise loss BPR is chosen, the optimization object will be

(8)

In terms of node classification, node embedding will be transformed to a probability distribution representing which class it belongs to, shown as follows,

(9)

where is the distribution and is the number of classes. Similarly, the training data will be , where and belongs to the class means ; otherwise . In general, the point-wise loss function is usually chosen for the classification task, such as cross-entropy loss which is formulated as

(10)

In short, optimization in GNN-based models treats representations generalized by GNNs as input and graph structures (e.g., edges, node classes) as labels, and loss functions are defined for training.

2.3. Why are GNNs required for recommender systems

Over the past decade, recommender systems have evolved rapidly from traditional factorization approaches to advanced deep neural networks based models. Particularly, GNN based recommenders have achieved the state-of-the-art in many aspects, including different recommendaiton stages, scenarios, objectives, applications. The success of GNN based recommenders can be explained from the following three perspectives: (1) structural data; (2) high-order connectivity; (3) supervision signal.

Structural data. Data collected from online platforms comes in many forms, including user-item interaction (rating, click, purchase, etc.), user profile (gender, age, income, etc.), item attribute (brand, category, price, etc.), and so on. Traditional recommender systems are not capable of leveraging those multiple forms of data, and they usually focus on one or a few specific data sources, which leads to sub-optimal performance since much information is ignored. By expressing all the data as nodes and edges on a graph, GNN provides a unified way to utilize available data. Meanwhile, GNN shows strong power in learning representations, and thus, high-quality embeddings for users, items, and other features can be obtained, which is critical to the recommendation performance.

High-order connectivity. Recommendation accuracy relies on capturing similarity between users and items, and such similarity is supposed to be reflected in the learned embedding space. Specifically, the learned embedding for a user is similar to embeddings of items that are interacted by the user. Furthermore, those items that are interacted by other users with similar preferences are also relevant to the user, which is known as the collaborative filtering effect, and it is of great importance for recommendation accuracy. In traditional approaches, the collaborative filtering effect is only implicitly captured since the training data is mainly interaction records that only contain directly connected items. In other words, only first-order connectivity is taken into consideration. The absence of high-order connectivity can largely damage recommendation performance. In contrast, GNN based models can effectively capture high-order connectivity. Specifically, the collaborative filtering effect can be naturally expressed as multi-hop neighbors on the graph, and it is incorporated in the learned representations through embedding propagation and aggregation.

Supervision signal. Supervision signals are usually sparse in the collected data, while GNN based model can leverage semi-supervised signals in the representation learning process to alleviate this problem. Take the E-Commerce platform as an example; the target behavior, purchase, is pretty sparse compared to other behaviors. Therefore, recommender systems that only use the target behavior may achieve poor performance. GNN based models can effectively incorporate multiple non-target behaviors, such as search and add to cart, by encoding semi-supervised signals over the graph, which can significantly improve recommendation performance (Jin et al., 2020). Meanwhile, self-supervised signals can also be utilized by designing auxiliary tasks on the graph, which further improves recommendation performance.

3. Challenges of applying GNNs to recommender systems

Although it is well motivated to apply graph neural networks in recommender systems, there exist four parts of critical challenges.

  • How to construct appropriate graphs for specific tasks?

  • How to design the mechanism of information propagation and aggregation?

  • How to optimize the model?

  • How to ensure the efficiency of model training and inference?

In the following, we will elaborate on the four challenges one by one.

3.1. Graph Constrcution

Obviously, the first step of applying graph neural networks is to construct graphs. This is in two folds: constructing the data input as graph-structured data; reorganize the recommendation goal as a task on the graph. Taking the task of standard collaborative filtering as an example, the data input is the observed user-item interaction data, and the output is predictions of the missing user-item interactions. Therefore, a bipartite graph with users/items as nodes and interactions as edges can be constructed. Besides, the CF task turns to the user-item link prediction on the graph.

However, it is challenging to construct graphs that can well handle the task properly. It should be carefully implemented with the consideration of the following aspects.

  • [leftmargin=*]

  • Nodes. One of the main goals of learning with graph neural networks is to assign nodes the representations. This results in that the definition of nodes largely determines the scale of the GNN models, of which the majority of parameters are occupied by the layer- node embeddings. Note that edge embeddings are usually either not considered or computed based on node embeddings. On the other hand, it is also a challenging problem to determine whether to distinguish different types of nodes. For example, in the collaborative filtering task, user nodes and item nodes can be modeling differently or considered as the same kind of nodes. Another challenging point is to handle concrete input such as some numerical features like item prices, which are always continuous numbers. To represent these features in the graph, one possible solution is to discretize them to categorical ones, which can be then represented as nodes (Zheng et al., 2020a).

  • Edges. The definition of edges highly affects the graph’s quality in further propagation and aggregation, along with the model optimization. In some trivial tasks, the data input of the recommender system can be considered as a kind of relational data such as user-item interactions or user-user social relations. In some complex tasks, other relations can also be represented as edges. For example, in bundle recommendation, a bundle consists of several items. Then the edge connecting the bundle and item can reflect the relation of affiliation. Good designs of edges when constructing a graph should fully consider the graph density. A too-dense graph means there are nodes with extremely high degrees. This will make the embedding propagation conducted by a very large number of neighbors. It will further make the propagated embedding non-distinguished and useless. To handle too dense edges, sampling, filtering, or pruning on graphs are promising solutions. A too-sparse graph will also, of course, results in the poor utility of embedding propagation since the propagation will be conducted on only a small fraction of nodes.

3.2. Network Design

What makes GNN different from the traditional graph learning methods is the propagation layer. As for the propagation, how to choose the path is critical for modeling the high-order similarity in recommender systems. Besides, the propagation can also be parametric, which assigns different weights to different nodes. For example, when propagating item embeddings to a user node in the user-item interaction graph, it captures the item-based CF effect. The weights refer to the different importance of historically interacted items.

In the propagation, there are also various choices of aggregation functions, including mean pooling, LSTM, max, min, etc. Since there is no single choice that can perform the best among all recommendation tasks or different datasets, it is vital to design a specific and proper one. Besides, the different choices of propagation/aggregation highly affect the computation efficiency. For example, mean pooling is widely used in GNN-based recommendation models since it can be computed efficiently, especially for the graph containing high-degree nodes, such as very popular items (which can connect a large number of users). Also, the propagation/aggregation layers can be stacked to help nodes access higher-hops neighbors. Too shallow layers make the high-order graph structure cannot be well modeled, and too deep ones make the node embedding over-smoothed. Either one of the two cases will lead to poor recommendation performance.

3.3. Model Optimization

To optimize the graph neural network-based recommendation models, the traditional loss functions in recommender system always turn to graph learning losses. For example, the logloss in the optimization can be regarded as the point-wise link prediction loss. Similarly, BPR loss (Rendle et al., 2009) is usually adopted in the link prediction task on graphs. Another aspect is data sampling. In GNN-based recommendation, to sample positive or negative items, the sampling manner can highly depend on the graph structure. For example, in social recommendation, performing random walk on the graph can generate weak positive items (such as items interacted by friends).

In addition, sometimes, GNN-based recommendation may involve multiple tasks, such as the link prediction tasks on different types of edges. Then in such a case, how to balance each task and make them enhance each other is challenging.

3.4. Computation Efficiency

In the real world, recommender systems should be trained/inferred efficiently. Therefore, to ensure the application value of GNN-based recommendation models, its computation efficiency should be seriously considered. Compared with traditional non-GNN recommendation methods such as NCF or FM, GNN models’ computation cost is far higher. Especially for the spectral GNN models such as GCN, complex matrix operations are involved in each GCN layer. With multi-layer stacking of GCN layers, the computation cost further increases. Therefore, spatial GNN models such as PinSage can be easier to be implemented in large-scale industrial applications. With sampling among neighbors or pruned graph structure, efficiency can always be kept as long as we can bear the drop of recommendation performance.

4. Existing Methods

4.1. Taxonomy 

In recent years, GNN has been applied to a wide range of recommendation tasks. Here we define the taxonomy in terms of recommendation stages, scenarios, objectives, and applications, respectively. To be more specific, recommendation stages indicate the overall procedure that a recommender system is implemented in the real-world platform. The procedure includes matching for item candidates selection, ranking for capturing user preferences, and re-ranking for other criteria beyond accuracy. The recommendation scenarios including social recommendation, sequential recommendation, cross-domain recommendation, etc. The recommendation objectives incorporate accuracy, diversity, explainability, fairness, and so on, in which accuracy is of the most concern. The recommendation application refers to specific industrial applications. Table 2, 3, and 4 show representative researches of GNN-based recommendation published in top-tier venues for different recommendation stages, recommendation scenarios, and recommendation objectives, respectively.

Perspective Category Model Venue Year
Stage Matching GCMC (Berg et al., 2018) KDD 2018
PinSage (Ying et al., 2018) KDD 2018
NGCF (Wang et al., 2019b) SIGIR 2019
LightGCN (He et al., 2020) SIGIR 2020
Ranking Fi-GNN (Li et al., 2019a) CIKM 2019
PUP (Zheng et al., 2020a) ICDE 2020
-SIGN (Su et al., 2021) AAAI 2021
DG-ENN (Guo et al., 2021a) KDD 2021
Re-ranking IRGPR (Liu et al., 2020d) CIKM 2020
Table 2. A summary of GNN-based models in different recommendation stages in top-tier venues.
Perspective Category Model Venue Year
Scenario Social DiffNet (Wu et al., 2019a) SIGIR 2019
GraphRec (Fan et al., 2019a) WWW 2019
DANSER (Wu et al., 2019c) WWW 2019
DGRec (Song et al., 2019b) WSDM 2019
HGP (Kim et al., 2019) RecSys 2019
DiffNet++ (Wu et al., 2020b) TKDE 2020
MHCN (Yu et al., 2021b) WWW 2021
SEPT (Yu et al., 2021a) KDD 2021
GBGCN (Zhang et al., 2021a) ICDE 2021
KCGN (Huang et al., 2021) AAAI 2021
DiffNetLG (Song et al., 2021) SIGIR 2021
Sequential ISSR (Liu et al., 2020c) AAAI 2020
MA-GNN (Ma et al., 2020b) AAAI 2020
STP-UDGAT(Lim et al., 2020) CIKM 2020
GPR(Chang et al., 2020b) CIKM 2020
GES-SASRec(Zhu et al., 2021) TKDE 2021
RetaGNN(Hsu and Li, 2021) WWW 2021
TGSRec(Fan et al., 2021) CIKM 2021
SGRec(Li et al., 2021b) IJCAI 2021
SURGE (Chang et al., 2021) SIGIR 2021
Session SR-GNN (Wu et al., 2019b) AAAI 2019
GC-SAN (Xu et al., 2019c) IJCAI 2019
TA-GNN (Yu et al., 2020b) SIGIR 2020
MGNN-SPred (Wang et al., 2020c) WWW 2020
LESSR (Chen and Wong, 2020) KDD 2020
MKM-SR (Meng et al., 2020) SIGIR 2020
GAG (Qiu et al., 2020) SIGIR 2020
GCE-GNN (Wang et al., 2020b) SIGIR 2020
SGNN-HN (Pan et al., 2020) CIKM 2020
DHCN (Xia et al., 2021e) AAAI 2021
SHARE (Wang et al., 2021a) SDM 2021
SERec (Chen and Wong, 2021) WSDM 2021
COTREC (Xia et al., 2021d) CIKM 2021
DAT-MID (Chen et al., 2021a) SIGIR 2021
TASRec (Zhou et al., 2021) SIGIR 2021
Bundle BGCN(Chang et al., 2020a) SIGIR 2020
HFGN(Li et al., 2020c) SIGIR 2020
BundleNet(Deng et al., 2020) CIKM 2020
DPR(Zheng et al., 2021c) WWW 2021
Cross Domain PPGN(Zhao et al., 2019b) CIKM 2019
BiTGCF(Liu et al., 2020a) CIKM 2020
DAN(Wang et al., 2020d) CIKM 2020
HeroGRAPH(Cui et al., 2020) Recsys 2020
DAGCN(Guo et al., 2021b) IJCAI 2021
Table 3. A summary of GNN-based models in different recommendation scenarios in top-tier venues.
Perspective Category Model Venue Year
Objective Multi-behavior MBGCN (Jin et al., 2020) SIGIR 2020
MGNN-SPred(Wang et al., 2020c) WWW 2020
MGNN(Zhang et al., 2020) CIKM 2020
LP-MRGNN(Wang et al., 2021b) TKDE 2021
GNMR(Xia et al., 2021a) ICDE 2021
MB-GMN(Xia et al., 2021c) SIGIR 2021
KHGT(Xia et al., 2021b) AAAI 2021
GHCF(Chen et al., 2021c) AAAI 2021
DMBGN(Xiao et al., 2021) KDD 2021
Diversity V2HT (Li et al., 2019b) CIKM 2019
BGCF (Sun et al., 2020a) KDD 2020
DGCN (Zheng et al., 2021a) WWW 2021
Explainability RippleNet (Wang et al., 2018) CIKM 2018
EIUM (Huang et al., 2019) MM 2019
KPRN (Wang et al., 2019c) AAAI 2019
RuleRec (Ma et al., 2019) WWW 2019
PGPR (Xian et al., 2019) SIGIR 2019
KGAT (Wang et al., 2019a) KDD 2019
TMER (Chen et al., 2021b) WSDM 2021
Fairness FairGo (Wu et al., 2021a) WWW 2021
FairGNN (Dai and Wang, 2021) WSDM 2021
Table 4. A summary of GNN-based models for different recommendation objectives in top-tier venues.

4.2. GNN in Different Recommendation Stages

4.2.1. GNN in Matching

In the matching stage, efficiency is an essential problem because of the high computation complexity for candidates selection. Specifically, only hundreds of items will be selected from the item pool of million magnitudes for the following ranking stage, based on coarse-grained user preferences. Therefore, proposed models in this stage barely leverage user-item interactions as data input for modeling user preferences, without introducing additional features such as user ages, item price, browsing time on the application, etc.

GNN-based models in the matching stage can be regarded as embedding matching, usually designing specialized GNN architecture on the user-item bipartite graph (Berg et al., 2018; Wang et al., 2019b; Sun et al., 2020b; Wang et al., 2020a; Wu et al., 2021c). Berg et al. (Berg et al., 2018) proposed to pass neighborhood messages by summing and assign weight-sharing transformation channels for different relational edges (i.e., user-item ratings). Wang et al. (Wang et al., 2019b) proposed a spatial GNN in recommendation and obtain superior performance compared with conventional CF methods like MF (Koren et al., 2009) or NCF (He et al., 2017). Sun et al. (Sun et al., 2020b) argued that simple aggregation mechanisms like sum, mean, or max, cannot model relational information among neighbors and proposed a neighbor interaction-aware convolution to address the issue. Wang et al. (Wang et al., 2020a) developed disentangled GNN to capture independent user intentions, which extends the set of candidate items in matching and guarantees the accuracy simultaneously. Wu et al. (Wu et al., 2021c) leverage the stability of graph structure to incorporate a contrastive learning framework to assist representation learning. These GNN-based models are capable of capturing high-order similarity among users and items as well as structural connectivity. In this way, the semantics that users with similar interactions will have similar preferences are extended through multiple times of information propagation. On the other hand, the training complexity of GNN-based models were demonstrated (Wang et al., 2019b, 2020a) acceptable and comparable with non-graph models, especially when the transformation matrix is removed (He et al., 2020). Besides, (Ying et al., 2018) showed that GNN-based model could be applied to web-scale recommender system in real-world platforms efficiently and effectively, which combines random walk and GraphSAGE (Hamilton et al., 2017) for embedding learning on a large-scale item-item graph. Table 5 shows the commonality and difference among the GNN models in matching.

In a nutshell, GNN can be applied to recommendation tasks effectively, which balances the accuracy and efficiency of generating candidates from the item pool.

4.2.2. GNN in Ranking

In the ranking stage, with a much smaller amount of candidate items, more accurate models can be utilized, and more features can be included. Existing ranking models usually first transform sparse features into one-hot encodings then transform them into dense embedding vectors. These feature embeddings are directly concatenated and fed into DNN 

(Guo et al., 2017; Cheng et al., 2016) or specifically designed models (Rendle, 2010; He et al., 2017; Song et al., 2019a) in an unstructured way to estimate the ranking score. The main challenge of utilizing GNN for ranking is designing proper structures to capture feature interactions. Specifically, GNN-based ranking models usually consist of two components, encoder and predictor, which address feature interaction from different directions. On the one hand, special graph structures can be designed in the encoder to capture the desired feature interactions. On the other hand, feature interaction can be taken into consideration in the predictor, where the ranking score is estimated by integrating different feature embeddings from the GNN encoder.

Li et al. (Li et al., 2019a) propose Feature Interaction Graph Neural Networks (Fi-GNN), which constructs a weighted fully connected graph of all the input features. The encoder in Fi-GNN is composed of a GAT and a GRU, and the predictor is achieved with attention networks. Zheng et al. (Zheng et al., 2020a, 2021b) investigate the influence of price feature in ranking and propose a model called Price-aware User Preference modeling (PUP). They design an encoder with GCN on a pre-defined heterogeneous graph to capture price awareness, and a two-branch factorization machine is utilized as the predictor. Since not all feature interactions are useful, -SIGN (Su et al., 2021) automatically detects beneficial feature interactions and only reserves those edges, resulting in a learned graph which is further fed into a graph classification model to estimate the ranking score. In addition, Guo et al. (Guo et al., 2021a) propose DG-ENN with a dual graph of an attribute graph and a collaborative graph, which integrates the information of different fields to refine the embedding for ranking. Furthermore, SHCF (Li et al., 2021d) and GCM (Wu et al., 2020a) utilize extra nodes and edge attributes to represent item attributes and context information, respectively. Classical interaction predictors are adopted, such as inner product and FM. Table 6 illustrates the differences of the above GNN based ranking models in terms of the designs of encoder and predictor.

Model Graph GNN Motivations
GCMC (Berg et al., 2018) user-item GCN weight sharing among relations
NGCF (Wang et al., 2019b) user-item GCN -
DGCF (Wang et al., 2020a) user-item GAT disentangled representations
LightGCN (He et al., 2020) user-item LightGCN remove transformation and nonlinearity
SGL (Wu et al., 2021c) user-item GCN self-supervision on graphs
NIA-GCN (Sun et al., 2020b) user-item NIA-GCN neighbor interaction (NI)
PinSage (Ying et al., 2018) item-item GCN sampling
Table 5. Details of GNN models in matching stage.
Model Graph GNN Predictor
Fi-GNN (Li et al., 2019a) fully-connected GAT+GRU attention
PUP (Zheng et al., 2020a) pre-defined GCN two-branch FM
-SIGN (Su et al., 2021) learned SIGN graph classification
DG-ENN (Guo et al., 2021a) pre-defined LightGCN DNN
SHCF (Li et al., 2021d) pre-defined HGAT inner product
GCM (Wu et al., 2020a) pre-defined context GCN FM
Table 6. Details of GNN models in ranking stage.

4.2.3. GNN in Re-ranking

After obtaining the scores of recommended items, top items are further re-ranked with pre-defined rules or functions to improve the recommendation quality. Specifically, two key factors need to be considered in re-ranking. First, different items can have mutual influence by certain relationships such as substitutability and complementarity. Second, different users tend to have distinct preferences, and thus re-ranking can also be personalized. GNN provides a unified way to encode both item relationships and user preferences. Liu et al. (Liu et al., 2020d) propose a model called IRGPR to accomplish personalized re-ranking with the help of GNN. They propose a heterogeneous graph to fuse the two information sources, one item relation graph to capture multiple item relationships, and one user-item scoring graph to include the initial ranking scores. User and item embeddings are obtained after multiple message propagation layers, including global item relation propagation and personalized intent propagation. The final order of re-ranked items is generated with a feed-forward network.

4.3. GNN in Different Recommendation Scenarios

4.3.1. GNN in Social Recommendation

In social recommendation, we have social networks that contain the social relations of each user, and the goal is to utilize the local neighbors’ preferences for each user in social networks to enhance the user modeling (Wu et al., 2018a, b; Sun et al., 2018a; Chen et al., 2019b). From the perspective of representation learning with GNN, there are two key considerations in social recommendation: 1) how to capture the social factor; 2) how to combine the social factor from friends and user preference from his/her interaction behaviors. Here, we first summarize how the existing works capture the social factors from two perspectives, i.e., graph construction, and information propagation.

  • [leftmargin=*]

  • Graph construction. In social-aware recommender systems, user’s final behavior is decided by both the social impacts from friends and his/her own preferences. One of the main challenges in social recommendation is how to construct a social graph to capture the social influences from friends. Generally speaking, a certain user in social networks is not only influenced by his/her friends (the first-order neighbors) but also influenced by friends’ friends (the high-order neighbors). To capture the high-order social relations, the graph construction methods can be divided into two directions: stacked graphs and hypergraph. Given that normal graph can only model the pairwise relations, normal graph-based methods (Fan et al., 2019a; Wu et al., 2019a, 2020b, c; Yu et al., 2021a, 2020a; Guo and Wang, 2020; Luo et al., 2020a; Xu et al., 2020; Kim et al., 2019; Mu et al., 2019; Song et al., 2021) stacked multiple GNN layers to capture multi-hop high-order social relations. However, stacked GNN layers may suffer from the over-smoothing (Chen et al., 2020a) problem, which may lead to significant performance degradation. Hypergraph-based methods, such as MHCN (Yu et al., 2021b), propose to model the high-order social relations with hyperedge (Feng et al., 2019), which can connect more than two nodes and model the high-order relations in a natural way. HOSR (Liu et al., 2020b) recursively propagate embeddings along the social network for reflecting the influence of high-order neighbors in the user representations. To further improve the recommendation performance, some works (Xu et al., 2019a; Zhang et al., 2021a; Song et al., 2019b; Huang et al., 2021; Bai et al., 2020) introduce side information when constructing the graph. RecoGCN (Xu et al., 2019a) unifies users, items, and selling agents into a heterogeneous graph to capture the complex relations in social E-commerce. GBGCN (Zhang et al., 2021a) constructs a graph for organizing user behaviors of two views in group-buying recommendation, of which the initiator view contains those initiator-item interactions and the participant view contains those participant-item interactions. DGRec (Song et al., 2019b) and TGRec (Bai et al., 2020) introduce temporal information of user behaviors into social recommendation. KCGN (Huang et al., 2021) proposes to capture both user-user and item-item relations with the developed knowledge-aware coupled graph.

    Figure 12. Illustration of GNN models for social recommendation.
  • Information propagation. As for the propagation on the constructed graph for social recommendation, there are two main propagation mechanisms, i.e., average-pooling (GCN) and attention mechanism (GAT). The methods with average-pooling mechanism (Wu et al., 2019a; Yu et al., 2021b, a; Guo and Wang, 2020; Zhang et al., 2021a; Xu et al., 2020; Huang et al., 2021; Kim et al., 2019; Song et al., 2021; Bai et al., 2020) conduct average-pooling propagation (GCN) on social graph and treats the social influence of friends equally. RecoGCN (Xu et al., 2019a) conducts meta-path based GCN propagation on the constructed graph to capture both the social impact and user preference. HOSR (Liu et al., 2020b) aggregates the information from neighbors with GCN to capture the high-order relations in social graph. MHCN (Yu et al., 2021b) performs propagation with GCN on constructed hypergraph to obtain the high-order social relations. The methods with attention mechanism (Fan et al., 2019a; Wu et al., 2020b, 2019c; Yu et al., 2020a; Luo et al., 2020a; Song et al., 2019b; Mu et al., 2019), such as GraphRec (Fan et al., 2019a) and DiffNet++ (Wu et al., 2020b), assume that the social influences from different neighbors on the social graph are different and assign different weights to the social influences from different friends.

Model Graph GNN Social Signal Extraction
DiffNet (Wu et al., 2019a) social graph GCN sum-pooling
GraphRec (Fan et al., 2019a) social graph + user-item graph GAT concatenation
DANSER (Wu et al., 2019c) social graph + user-item graph GAT & GCN -
DiffNet++ (Wu et al., 2020b) heterogeneous graph GAT multi-level attention network
MHCN (Yu et al., 2021b) multi-channel hypergraph + user-item graph HyperGCN sum-pooling
SEPT (Yu et al., 2021a) triangle-graphs + user-item graph GCN -
RecoGCN (Xu et al., 2019a) heterogeneous graph Meta-path + GCN concatenation
ESRF (Yu et al., 2020a) motif-induced graph GAT sum-pooling
GNN-SoR (Guo and Wang, 2020) heterogeneous graph GCN concatenation
ASR (Luo et al., 2020a) heterogeneous graph GAT concatenation
GBGCN (Zhang et al., 2021a) heterogeneous graph GCN -
DGRec (Song et al., 2019b) social graph GAT -
SR-HGNN (Xu et al., 2020) social graph + user-item graph GCN concatenation
KCGN (Huang et al., 2021) social graph + item-item graph GCN concatenation
HGP (Kim et al., 2019) group-user graph + user-item graph GCN attention mechanism
GAT-NSR (Mu et al., 2019) social graph + user-item graph GAT MLP
HOSR (Liu et al., 2020b) social graph + user-item graph GCN attention mechanism
DiffNetLG (Song et al., 2021) heterogeneous graph GCN concatenation
TGRec (Bai et al., 2020) heterogeneous graph GCN attention mechanism
Table 7. Details of GNN models for social recommendation.

In social recommendation, user representations are learned from two distinct perspectives, i.e. social influence and user interactions. To combine the user representations from the above two perspectives, there are two strategies, 1) separately learn user representations from the social graph and user-item bipartite graph and 2) jointly learn user representations from the unified graph that consists of social graph and user-item bipartite graph. The methods with the first strategy, such as DiffNet (Wu et al., 2019a), GraphRec (Fan et al., 2019a) and MHCN (Yu et al., 2021b), first separately learn user representations from social graph and user-item graph, and then combines the representations with sum-pooling (Wu et al., 2019a; Yu et al., 2021b), concatenation (Fan et al., 2019a), MLP (Mu et al., 2019) or attention mechanism (Kim et al., 2019; Liu et al., 2020b; Bai et al., 2020). DiffNet++ (Wu et al., 2020b), a typical method with the second strategy, first aggregates the information in the user-item sub-graph and social sub-graph with the GAT mechanism and then combines the representations with the designed multi-level attention network at each layer. Table 7 shows the differences among the above approaches for social recommendation.

To sum up, the development of social recommendation with GNN can be summarized in Fig. 12. Early efforts in Social Recommendation only model the social network with GNN, such as DiffNet (Wu et al., 2019a). Then, the methods (Fan et al., 2019a; Wu et al., 2019a, 2020b, c; Yu et al., 2021a, 2020a; Guo and Wang, 2020; Luo et al., 2020a; Xu et al., 2020; Kim et al., 2019; Mu et al., 2019; Song et al., 2021) that model both social network and user interactions with GNNs become the mainstream in GNN-based social recommendation. Moreover, some studies, such as MHCN (Yu et al., 2021b) and HOSR (Liu et al., 2020b), attempt to enhance the recommendation by modeling the high-order relations in social networks more sufficiently. Also, there exist some works (Zhang et al., 2021a; Song et al., 2019b; Xu et al., 2019a; Bai et al., 2020; Huang et al., 2021) that introduce additional information to further enhance the social recommendation.

4.3.2. GNN in Sequential Recommendation

Figure 13. Illustration of GNN models for sequential recommendation.
Model Graph GNN Sequential Modeling
SURGE (Chang et al., 2021) item-item graph GAT RNN
(Wang and Cai, 2020) item-item graph GCN Attention
ISSR(Liu et al., 2020c) item-item and user-item graph GCN RNN
MA-GNN(Ma et al., 2020b) item-item graph GCN Memory network
DGSR(Zhang et al., 2021c) user-item graph GAT RNN
GES-SASRec(Zhu et al., 2021) item-item graph GCN RNN
RetaGNN(Hsu and Li, 2021) temporal heterogeneous graph GAT Self Attention
TGSRec(Fan et al., 2021) temporal user-item graph GAT GAT
SGRec(Li et al., 2021b) item-item graph GAT GAT
GME(Xie et al., 2016) item-item and user-item graph GAT GAT
STP-UDGAT(Lim et al., 2020) item-item and user-item graph GAT GAT
GPR(Chang et al., 2020b) item-item and user-item graph GCN GCN
Table 8. Details of GNN models for sequential recommendation.

For sequential recommendation, in order to improve the recommendation performance, it is necessary to extract as much effective information as possible from the sequence, and to learn the user’s interest in the sequence, including short-term interest, long-term interest, dynamic interest, etc., so as to accurately predict the next item that the user may be interested in. Some tools for sequence modeling have been used, such as Markov chains 

(Cheng et al., 2013)

or recurrent neural networks 

(Kang and McAuley, 2018). For graph neural networks, it can be well leveraged for short-term, dynamic interest modeling or representation learning in by converting the data to graph. A general pattern for sequential modeling with GNN is shown in figure 13.

SURGE (Chang et al., 2021) transforms the sequence of each user into an item-item graph and adaptively learns the weights of edges through metric learning, with only the stronger edges retained by dynamic graph pooling. The retained graph is converted to a sequence by position flatten and finally be used to predict the next item. Ma et al.  (Ma et al., 2020b) considers the short-term interest modeling in the sequence to build an item-item graph. For each item, it only builds edges with other items close to it in the sequence. This enables it to learn short-term user interests in the sequence while still learning long-term user interests through other networks. The learned multiple representations are fused together and used for final recommendation.

Since GNN has the ability of high-order relationship modeling by aggregating information from neighbor nodes, after fusing multiple sequences into one graph, it can learn representations of both users and items in different sequences, which can’t be accomplished by Markov model or recurrent neural network. Wang

et al. (Wang and Cai, 2020) propose a simple method that directly converts the sequence information into directed edges on the graph and then uses GNN to learn representations. Liu et al.  (Liu et al., 2020c) construct a user-item bipartite graph and an item-item graph at the same time, where the edges of the item-item graph indicate co-occurrence in a sequence, with edge weights assigned according to the number of occurrences. The representations learned by GNN are used in the final recommendation through the recurrent neural network. Different from directly converting the temporal sequence into directed edges in the graph, DGSR (Zhang et al., 2021c) and TGSRec (Fan et al., 2021) consider the timestamps in the sequence in the process of graph construction. In the graph, each edge represents the interaction between the user and the item, and has the corresponding time attribute. Then perform convolution operations on the temporal graph to learn the representations of users and items. GES-SASRec (Zhu et al., 2021) and SGRec (Li et al., 2021b) focus on the learning of item representations. For an item in a sequence, GES-SASRec (Zhu et al., 2021) considers the next item of this item in other sequences, and SGRec (Li et al., 2021b) not only considers the next item but also considers the previous one. By aggregating the items before and after the target item in different sequences, the representation of the item is enhanced. GPR (Chang et al., 2020b) and GME (Xie et al., 2016) constructs edges between items by considering the frequency of consecutive occurrences or occurrences in the same sequence to enhance the representation. Some works are more complicated. For example, RetaGNN (Hsu and Li, 2021) considers the attributes of the items when constructing the graph, while STP-UDGAT (Lim et al., 2020) considers the geographic location, timestamp, and frequency in the POI recommendation. Table 8 summarizes the above works.

4.3.3. GNN in Session-based Recommendation

Figure 14. Illustration of GNN models for session-based recommendation (SBR).
Model Graph GNN Enrich Graph Structure
SR-GNN (Wu et al., 2019b) directed graph gated GNN -
GC-SAN (Xu et al., 2019c) directed graph gated GNN -
TA-GNN (Yu et al., 2020b) directed graph gated GNN -
FGNN (Qiu et al., 2019) directed graph GAT -
A-PGNN (Wu et al., 2019d) directed graph gated GNN cross sessions
MGNN-SPred (Wang et al., 2020c) directed & multi-relational item graph GraphSAGE cross sessions
GAG (Qiu et al., 2020) directed graph GCN cross sessions
SGNN-HN (Pan et al., 2020) star graph gated GNN additional edges
GCE-GNN (Wang et al., 2020b) directed graph + global graph GAT cross sessions
DGTN (Zheng et al., 2020b) directed graph GCN cross sessions
DHCN (Xia et al., 2021e) hypergraph + line graph HyperGCN cross sessions
SHARE (Wang et al., 2021a) session hypergraph HyperGAT additional edges
SERec (Chen and Wong, 2021) KG + directed graph GAT + gated GNN cross sessions
LESSR (Chen and Wong, 2020) directed graph GAT additional edges
CAGE (Sheu and Li, 2020) KG + article-level graph GCN cross sessions
MKM-SR (Meng et al., 2020) KG + directed graph gated GNN cross sessions
COTREC (Xia et al., 2021d) item graph + line graph GCN cross sessions
DAT-MDI (Chen et al., 2021a) directed graph + global graph GAT cross sessions
TASRec (Zhou et al., 2021) dynamic graph GCN cross sessions
Table 9. Details of GNN models for session-based recommendation.

In session-based recommendation, the session data may contain both user interersts and noisy signals. Suppose a session for a certain user, iPhone iPad milk AirPods. Obviously, milk is likely clicked by mistake and then becomes a noise, as the session reflects the user’s preference for electronic products. Hence, the two main considerations in session-based recommendation are 1) how to model the item transition pattern in session data, and 2) how to activate user’s core interests from noisy data. From the perspective of graph learning, the item transitions can be modeled as graph, and the information propagation on the graph can activate user’s actual interests.

  • [leftmargin=*]

  • Graph construction. In session-based recommendation, most existing works (Wu et al., 2019b; Qiu et al., 2019; Wang et al., 2020b; Xu et al., 2019c; Yu et al., 2020b) model the session data with a directed graph to capture the item transition pattern. Distinct from sequential recommendation, the session sequence in session-based recommendation is short and the user behaviors are limited, i.e., the average length of sequences in Tmallhttps://www.tmall.com is only 6.69 (Wang et al., 2020b; Xia et al., 2021e). Hence, a session graph constructed from a single session may only contain limited nodes and edges. To address the above challenge and sufficiently capture the possible relations among items, there are two strategies, 1) straightforwardly capturing relations from other sessions, and 2) add the additional edges of the session graph. For the first strategy, A-PGNN (Wu et al., 2019d), DGTN (Zheng et al., 2020b), and GAG (Qiu et al., 2020) propose to enhance relations of the current session graph with related sessions, and GCE-GNN (Wang et al., 2020b) leverages the global context by constructing another global graph to assist the transition patterns in the current session. DHCN (Xia et al., 2021e) regards each session as a hyperedge and represents all sessions in a hypergraph to model the high-order item relations. SERec (Chen and Wong, 2021) enhances the global information for each session with a knowledge graph. CAGE (Sheu and Li, 2020) learns the representations of semantic-level entities by leveraging the open knowledge graph to improve the session-based news recommendation. MKM-SR (Meng et al., 2020) enhances the information in the given session by incorporating user micro-behaviors and item knowledge graph. COTREC (Xia et al., 2021d) unifies all sessions into a global item graph from item view and captures the relations among sessions by line graph from session view. DAT-MID (Chen et al., 2021a) follows GCE-GNN (Wang et al., 2020b) to construct both session graph and global graph and then learns item embeddings from different domains. TASRec (Zhou et al., 2021) constructs a graph for each day to model the relations among items and enhance the information in each session. As for the second strategy, SGNN-HN (Pan et al., 2020) constructs star graph with a ”star” node to gain extra knowledge in session data. SHARE (Wang et al., 2021a) expands the hyperedge connections by sliding the contextual window on the session sequence. LESSR (Chen and Wong, 2020) proposes to first construct an edge-order preserving multigraph and then construct a shortcut graph for each session for enriching edge links.

  • Information propagation. As for the information propagation on the constructed graph, there are four propagation mechanisms that are used in session-based recommendation, e.g. gated GNN, GCN, GAT, and GraphSAGE. SR-GNN (Wu et al., 2019b) and its related works (Xu et al., 2019c; Yu et al., 2020b; Pan et al., 2020; Wu et al., 2019d; Meng et al., 2020; Chen and Wong, 2021)

    combine the gated recurrent units in the propagation (gated GNN) on the session graph. GAG 

    (Qiu et al., 2020), DCTN (Zheng et al., 2020b) conduct graph convolution on the constructed directed graph. DHCN (Xia et al., 2021e) proposes to perform graph convolution on both hypergraph and line graph to obtain session representations from two different perspectives. Similar to DHCN (Xia et al., 2021e), COTREC (Xia et al., 2021d) performs GCN on item graph and line graph to obtain information from item and session views, respectively. CAGE (Sheu and Li, 2020) conducts GCN on article-level graph and TASRec (Zhou et al., 2021) performs graph convolution on the dynamic graph to capture the item relations. FGNN (Qiu et al., 2019) conducts GAT on a directed session graph to assign different weights on different items. SHARE (Wang et al., 2021a) performs GAT on session hypergraphs to capture the high-order contextual relations among items. GCE-GNN (Wang et al., 2020b) and DAT-MID (Chen et al., 2021a) perform GAT on both session graph and global graph to capture the local and global information, respectively. MGNN-SPred (Wang et al., 2020c) adopts GraphSAGE on multi-relational item graph to capture the information from different types of neighbors.

Table 9 shows the differences among the above approaches for session-based recommendation. To sum up, the development of session-based recommendation (SBR) with GNN can be summarized in Fig. 14. Early efforts in SBR only model each session sequence with a directed graph, such as SR-GNN (Wu et al., 2019b), GC-SAN (Xu et al., 2019c), TA-GNN (Yu et al., 2020b), and FGNN (Qiu et al., 2019). Then, some methods attempt to enrich the relation and information in the session graph with other sessions or additional links. The methods in (Wu et al., 2019d; Wang et al., 2020c; Qiu et al., 2020; Wang et al., 2020b; Zheng et al., 2020b; Xia et al., 2021e; Chen and Wong, 2021; Sheu and Li, 2020; Meng et al., 2020; Xia et al., 2021d; Chen et al., 2021a; Zhou et al., 2021) combine the information from other sessions to capture more information from similar sessions or all sessions. Moreover, the methods with additional links, such as SGNN-HN (Pan et al., 2020), SHARE (Wang et al., 2021a) and LESSR (Chen and Wong, 2020), attempt to introduce additional edges to the given session to capture the complex relations and information in session data. Furthermore, some studies, such as DHCN (Xia et al., 2021e) and SHARE (Wang et al., 2021a), attempt to enhance the recommendation by modeling the high-order relations in session data more sufficiently.

4.3.4. GNN in Bundle Recommendation

Figure 15. Illustration of GNN models for bundle recommendation.
Model C1:modeling affiliation C2: refining bundle C3: high-order relation
BGCN(Chang et al., 2020a) tripartite graph meta-path propagation multi-layer GCN
HFGN(Li et al., 2020c) hierarchical Graph direct aggregation multi-layer GCN
BundleNet(Deng et al., 2020) tripartite graph transformation parameters multi-layer GCN
DPR(Zheng et al., 2021c) tripartite graph graph induction multi-layer GCN
Table 10. Details of GNN models for bundle recommendation (how they address challenges).

The three challenges of bundle recommendation are 1) users’ decisions towards bundles are determined by the items the bundles contain (the affiliation relation), 2) learning bundle representations with the sparse user-bundle interactions, and 3) high-order relations. The earlier works approach the problem of bundle recommendation by learning from user-item interaction and user-bundle interaction together, with parameter sharing or joint loss function (Pathak et al., 2017; Chen et al., 2019a). For the first time, Chang et al. (Chang et al., 2020a) propose a GNN-based model that unifies both two parts of interactions and the bundle-item affiliation-relations into one graph. Then the item can serve as the bridge for embedding propagation between user-bundle and bundle-bundle. Besides, a specially designed sampling manner for finding hard-negative samples is further proposed for training. Deng et al. (Deng et al., 2020) construct a similar tripartite graph with the transformation parameters to well extract bundle representations from included items’ representations. Zheng et al. (Zheng et al., 2021c) consider the bundle recommendation in the drug package dataset and propose to initialize a graph with auxiliary data and represent the interaction as scalar weights and vectors. GNN layers are proposed for obtaining the drug package embeddings. Li et al. (Li et al., 2020c) consider the problem of personalized outfit recommendation, which can also be regarded as a kind of bundle recommendation. The authors construct a hierarchical graph, of which the users, items, and outfits are contained. GNN layers are deployed for obtaining the representations of users and outfits. The learning of the GNN model follows a multi-task manner.

In summary, the data input of bundle recommendation can be well represented as graph-structural data, especially for the bundles, which have been represented as a kind of new nodes.

4.3.5. GNN in Cross-Domain Recommendation

Figure 16. Illustration of GNN models for cross-domain recommendation.
Model Graph Information Transferring
PPGN(Zhao et al., 2019b) cross-domain graph cross-domain propagation
BiTGCF(Liu et al., 2020a) domain-specific graphs common user attributes
DAN(Wang et al., 2020d) domain-specific graphs two-branched decoder
HeroGRAPH(Cui et al., 2020) cross-domain graph cross-domain propagation
DAGCN(Guo et al., 2021b) cross-domain graph cross-domain propagation
Table 11. Details of GNN models of cross-domain recommendation.

Benefit from the powerful capabilities, GNN-based recommendation model has gradually emerged in the field of cross-domain recommendation. Zhao et al. (Zhao et al., 2019b) construct the cross-domain graph with shared users and items from multiple domains. The proposed PPGN’s embedding propagation layers can well learn the users’ preferences on multiple domains’ items under a multi-task learning framework. Liu et al. (Liu et al., 2020a) propose the bidirectional knowledge transfer by regarding the shared users as the bridge. GNN layers are adopted to leverage the high-order connectivity in the user-item interaction graph for better preference learning with the user features, fuses from common features, and domain-specific features. Guo et al. (Guo et al., 2021b) construct the graph for each domain and deploy domain-specific GCN layers for learning user-specific embeddings. The authors combine it with the attention mechanisms for adoptively choose important neighbors during the embedding propagation. Wang et al. (Wang et al., 2020d) propose an encoder-decoder framework where the encoder is implemented by graph convolutional networks. Specifically, the GCNs are deployed on the user-item interaction graphs. Cui et al. (Cui et al., 2020) proposes to construct a heterogeneous graph, where multiple domains’ users and items can be well included. The GNN-based embedding propagation is deployed on multiple domains, where a user/item can directly absorb the information of different domains, where recurrent attention networks are used for distinguishing important neighbors.

4.3.6. GNN in Multi-behavior Recommendation

Figure 17. Illustration of GNN models for multi-behavior recommendation.
Model Graph GNN Multi-Behavior Modeling
MB-GMN(Xia et al., 2021c) user-item graph GCN / GAT Behavior representation
KHGT(Xia et al., 2021b) user-item / item-item graph GAT Weight
MBGCN(Jin et al., 2020) user-item graph GAT Weight
MGNN-SPred(Wang et al., 2020c) item-item graph GCN Weight
MGNN(Zhang et al., 2020) user-item graph GCN Node representation
LP-MRGNN(Wang et al., 2021b) item-item graph GCN Weight
GNNH(Yu et al., 2021d) item-item / category-category graph GCN Item representation
GNMR(Xia et al., 2021a) user-item graph GAT Weight
DMBGN(Xiao et al., 2021) item-item graph GCN Graph representation
GHCF(Chen et al., 2021c) user-item graph GCN Edge embedding
Table 12. Details of GNN models for multi-behavior recommendation.

Multiple types of behaviors can provide a large amount of information to the recommender system, which helps the recommender system to learn the user’s intentions better, thereby improving the recommendation performance. For recommender systems based on graph neural networks, on the basis of common user-item bipartite graphs, the multiple types of behaviors between users and items can naturally be modeled as different types of edges between nodes. Therefore, most of the multi-behavior recommendation methods based on graph neural networks are based on heterogeneous graphs. However, the focus of multi-behavior recommendation is 1) how to model the relationship between multiple behaviors and target behavior, and 2) how to model the semantics of the item through behavior, which is shown in Fig. 17.

In order to model the effect of auxiliary behaviors on target behaviors, the simplest method is to directly model all types of behaviors without considering the differences between behaviors. Zhang et al. (Zhang et al., 2020) constructs all user behaviors in one graph and performs graph convolution operations. Wang  (Wang et al., 2020c, 2021b) extracts each behavior from the graph to construct a subgraph, then learns from the subgraph, and finally aggregates through the gating mechanism. Chen et al.  (Chen et al., 2021c) proposed a intuitive method which assigns a representation to users, items, and behaviors. In the process of propagation, the representations of edges and neighbor nodes need to be composited first to obtain a new node representation, and then it can be applied to the GNN method. Through the composition operation, the representation of the node is fused with different types of behaviors. Xia et al.  (Xia et al., 2021a) also redesign the aggregation mechanism on the graph convolutional network to explicitly model the impact of different types of behavior. Jin et al.  (Jin et al., 2020) assign different learnable weights to different edges to model the importance of the behaviors. In addition, in order to capture complex multi-behavior relationships, there are also some works that rely on knowledge. For example, Xia et al.  (Xia et al., 2021b) learns the representations in different behavior spaces, and then inject temporal contextual information into the representations to model the user’s behavior dynamics, and finally, through the attention mechanism, discriminate the most important relationships and behaviors for the predicted target. Xia et al.  (Xia et al., 2021c) uses meta-graph networks to learn meta knowledge for different behaviors and then transfer the learned meta-knowledge between different types of behaviors.

In addition to modeling the influence of different types of behaviors, different behaviors may contain different meanings or semantics. For example, for items added to a shopping cart, users may have similar preferences for them, or these items have a complementary relationship and generally need to be purchased at the same time. If these items are connected through a graph, the representation of the item can be enhanced. In order to get better item representation, Yu et al. (Yu et al., 2021d) not only connects related items in the graph, but also constructs a new graph of the category that the items belong to, which is used to enhance the representation of the item. It needs to construct the item graph separately, and there are also some works that do not construct the graph separately and directly in the user-item heterogeneous graph, by using for meta-path or second-order neighbors, where similar items are aggregated to enhance their representation (Jin et al., 2020; Xia et al., 2021b). These works’ details are presented in Table 12.

4.4. GNN for Different Recommendation Objectives

4.4.1. GNN for Diversity

With respect to individual-level diversity, retrieved items from recommender systems need to cover more topics such as different categories of products or different genres of music. Therefore, utilizing GNN to increase diversity requires that the learned user embeddings should be close to item embeddings with various topics. However, as the embedding aggregation operation in GNN makes user embeddings close to embeddings of items that are interacted in historical records, GNN might discourage diversity by recommending too many similar items that belong to the dominant topic in users’ interaction history. For example, the learned embedding from GNN for a user who mainly interacts with electronics may be too close to embeddings of electronic items, making the GNN only recommend electronics to this user, which leads to low diversity. Therefore, in order to overcome the first challenge of weak signals from disadvantaged topics, efforts have been made to restrict the importance of dominant topics (electronics in the above example) by constructing diversified sub-graphs from the original user-item bipartite graph. Specifically, Sun et al. (Sun et al., 2020a) propose a model called Bayesian Graph Collaborative Filtering (BGCF), which constructs augmented graphs with node copying (Pal et al., 2019) from high-order neighbors, such that items of diverse topics with high similarity can be directly connected to user nodes. Zheng et al. (Zheng et al., 2021a) propose Diversified Graph Convolutional Networks (DGCN) and conduct rebalanced neighbor sampling, which down-weights dominant topics and boosts the importance of disadvantaged topics in neighbor nodes. Fig. 18 illustrates the comparison between BGCF and DGCN on graph construction for diversity. Meanwhile, to address the second challenge of balancing between accuracy and diversity, BGCF re-ranks the top items according to item popularity, and DGCN utilizes adversarial learning on the item embeddings which encourages GNN to capture user preferences that are largely independent with item categories, which further improves recommendation diversity.

Figure 18. Illustration of BGCF (Sun et al., 2020a) and DGCN (Zheng et al., 2021a) on addressing the challenge of weak preference signals.
Model C1: Weak Preference Signals C2: Accuracy-Diversity Balance
BGCF (Sun et al., 2020a) Node Copying Re-rank
DGCN (Zheng et al., 2021a) Neighbor Sampling Adversarial training
V2HT (Li et al., 2019b) Utilize item correlations -
FH-HAT (Xie et al., 2021) Heterogeneous graph Diversity loss
(Isufi et al., 2021) NN & FN graph Joint training
Table 13. Details of GNN models for diversified recommendation (how they address challenges)

As for system-level diversity, the main target is to discover more relevant items from the long-tail ones, which have much fewer training samples compared with those popular items. To address the weak signals of long-tail items, Li et al. (Li et al., 2019b) propose a model called V2HT to construct an item graph that explores item correlations with external knowledge. Specifically, four types of edges are introduced, which connect frequent items and long-tail items. Then multiple GCN layers are stacked, which propagates well-trained embeddings of frequent items to undertrained embeddings of long-tail items. In this way, long-tail item embeddings of higher quality are obtained since they share the information from frequent items; thus system-level diversity is improved with more recommendation on long-tail items.

In addition, a few studies (Xie et al., 2021; Isufi et al., 2021) utilize GNN to improve both individual-level and system-level diversity. Specifically, Xie et al. (Xie et al., 2021) propose FH-GAT, which addresses the challenge of weak signals by constructing a heterogeneous interaction graph to express diverse user preferences. A neighbor similarity-based loss is conducted on the heterogeneous graph to balance between accuracy and diversity. Isufi et al. (Isufi et al., 2021) propose two GCN on the nearest neighbor (NN) graph and the furthest neighbor (FN) graph, where NN guarantees accuracy and FN enhances the weak signals of diverse items. Meanwhile, the two GCN are jointly optimized with a hyper-parameter to achieves a trade-off between accuracy and diversity. Table 13 shows the differences among the above approaches.

4.4.2. GNN for Explainablility

Figure 19. Illusration of GNN models for explainable recommendation
Model GNN Source of Explainability
ECFKG (Ai et al., 2018) Meta-path based GE Meta-path over knowledge graph
RippleNet (Wang et al., 2018) Meta-path based GraphSage Meta-path over knowledge graph
EIUM (Huang et al., 2019) Meta-path based GE Meta-path over knowledge graph
KPRN (Wang et al., 2019c) Meta-path based GE Meta-path over knowledge graph
TMER (Chen et al., 2021b) Meta-path based GE Meta-path over knowledge graph and temporal dependency
RuleRec (Ma et al., 2019) Meta-path based GE Discovered meta-path over knowledge graph
PGPR (Xian et al., 2019) Meta-path based GE Discovered meta-path over knowledge graph
KGAT (Wang et al., 2019a) GAT Attention mechanism
HAGERec (Yang and Dong, 2020) GAT Attention mechanism
Table 14. Details of GNN models for explainable recommendation (how they achieve explainable recommendation)

With the proliferation of GNN, researchers also make endeavors to improve the explainability of recommender systems with GNN’s power of modeling logical relations. He et al. (He et al., 2015) construct a heterogeneous graph with three kinds of nodes, including the user, the item, and the aspect (the specific item property extracted from textual reviews). Therefore, they cast the recommendation task into a ternary relation ranking task, and propose TriRank with a high degree of explainability by explicitly modeling aspects in reviews.

Inspired by this work, the following research further explores rich information in dimensions of users and items, generally organized in a form of knowledge graph, in order to enhance the explainability (Zhang and Chen, 2020). Ai et al. (Ai et al., 2018) construct a knowledge graph with entities, i.e., users and items, and relations, e.g., “User A purchased Item B belonging to Category C”. Moreover, they embed each entity for recommendation and adopt the shortest relation path between a user and an item in the knowledge graph to indicate the recommendation explanations. Different from separately utilizing embedding-based and path-based methods like Ai et al. (Ai et al., 2018), Wang et al. (Wang et al., 2018) proposed an end-to-end framework RippleNet which combines the two knowledge graph-aware recommendation methods together. Here, the knowledge graph contains the related knowledge of the recommended items, for example, the type and author of a movie. In this way, explanations can be generated by the path between users’ history to an item with high scores.

The meta-path-based utilization of knowledge graph can also benefit other specific recommendation tasks, e.g., sequential recommendation (Huang et al., 2019; Wang et al., 2019c). Huang et al. (Huang et al., 2019) extract semantic meta-paths between a user and an item from knowledge graph to help sequential recommendation. Further, they encode and rank the meta-paths to generate the recommendation list, and these meta-paths also indicate the respective explanation. Similarly, Wang et al. (Wang et al., 2019c) also leverages knowledge graph to improve the performance of the sequential recommendation task and encode the meta-path between a user and an item with recurrent neural networks. Chen et al. (Chen et al., 2021b) further model the temporal meta-paths by capturing historical item features and path-defined context with neural networks.

However, these meta-path-based solutions also face some challenges in terms of how to obtain these meta-paths (Ma et al., 2019; Xian et al., 2019). First, considering predefined meta-paths requests extensive domain knowledge, Ma et al. (Ma et al., 2019) jointly combine the discovery of inductive rules (meta-paths) from the item-centric knowledge graph, which equips the framework with explainability and the learning of a rule-guided recommendation model. Moreover, to overcome the computational difficulties of enumerating all potential meta-paths, Xian et al. (Xian et al., 2019) replace the enumeration method with the reinforcement reasoning approach to identify proper meta-paths for scalability.

Besides meta-path-based solutions, Wang et al. (Wang et al., 2019a) propose a new method named Knowledge Graph Attention Network (KGAT), where the attention mechanism can offer explainability to some extent. Yang et al. (Yang and Dong, 2020) develop a hierarchical attention graph convolutional network to model higher-order relations in the heterogeneous knowledge graph, where the explainability is also dependent on the attention mechanism.

4.4.3. GNN for Fairness

Figure 20. Illustration of GNN models for fair recommendation.
Model GNN/GE Methods Enhancing Fairness
Fairwalk (Rahman et al., 2019) Node2vec (GE) learning fair graph embeddings by sampling
CFCGE (Bose and Hamilton, 2019) Invariant GE learning fair graph embeddings by adversarial learning
FairGo (Wu et al., 2021a) HetGNN(GNN) learning fair graph embeddings by adversarial learning
FairGNN (Dai and Wang, 2021) GraphSage(GNN)

learning fair GNN classifiers by adversarial learning

Table 15. Details of GNN models for fair recommendation.

Despite the power of graph data in recommendation, it might inherit or even amplify discrimination and the societal bias in recommendation (Dai and Wang, 2021; Rahman et al., 2019; Wu et al., 2021a). Past research has proven that compared with models that only adopt node attributes, the user unfairness is magnified due to the utilization of graph structures (Dai and Wang, 2021).

To curb the fairness issue in recommendation, some researchers propose to learn fair graph embeddings (Rahman et al., 2019; Bose and Hamilton, 2019; Wu et al., 2021a). Rahman et al. (Rahman et al., 2019) extends the well-known graph embedding method, node2vec (Grover and Leskovec, 2016), to a more fair version, Fairwalk, which can generate more diverse network neighborhood representation for social recommendations by sampling the next node based on its sensitive attributes. Thus, all nodes’ sensitive attributes are indispensable. Bose et al. (Bose and Hamilton, 2019) proposes an adversarial framework to minimize the sensitive information in graph embeddings with a discriminator enforcing the fairness constraints. Moreover, considering the evaluations of fairness can be varying, fairness constraints are flexible according to the task. However, similar to Fairwalk (Rahman et al., 2019), all nodes’ sensitive attributes are required. Wu et al. (Wu et al., 2021a) learn fair embeddings for recommendation from a graph-based perspective. They propose FairGo, which adopts a graph-based adversarial learning method to map embeddings from any recommendation models into a sensitive-information-filtered space, therefore eliminating potential leakage of sensitive information from both original recommendation embeddings and user-centric graph structures.

Indeed, except for learning fair embeddings, Dai et al. (Dai and Wang, 2021) propose FairGNN, which learns fair GNN classifiers with limited known sensitive attributes in an adversarial learning paradigm with fairness constraints. Different from fair graph embeddings, fair GNN classifiers are to ensure node classification task (rather than graph embeddings) independent with sensitive data. Moreover, they develop GNN-based sensitive data estimators to overcome the issue of missing sensitive data in the real world.

4.5. GNN for Specific Recommendation Applications

Graph neural networks are also widely used to handle the specific challenges in different applications of recommender systems. As for e-commerce/product recommendation, most of the existing works have been introduced in the above sections. Li et al. (Li et al., 2020b) propose to stack multiple GNN modules and use a deterministic clustering algorithm to help improve the efficiency of GNN in large-scale e-commerce applications. Liu et al. (Liu et al., 2021) propose to leverage the topology of item-relations for building graph neural networks in e-commerce recommendation. As for the Point-of-Interest recommendation, GGLR (Chang et al., 2020b) uses the sequence of POIs a user visited to construct the POI-POI graph, where the edges between POIs denote the frequency of users consecutively visiting two POIs. Zhang et al. (Zhang et al., 2021b) propose to combine social networks and user-item interactions together, which deploys embedding aggregation on both social-connected users and interacted POIs. As for news recommendation, Hu et al. (Hu et al., 2020) propose to introduce the preference disentanglement into the user-news embedding propagation.

There are a lot of papers for specific applications, but they can be well categorized into and covered by the corresponding stages, scenarios, and objectives, and thus we omit them here.

5. Open Problems and Future Directions

5.1. Go Deeper

Due to the over-smoothing problem, more and more studies focus on properly increasing GNN’s layers to capture higher-order connectivity correlations on graphs as well as improve models’ performance (Lai et al., 2020; Zhou et al., 2020b; Chen et al., 2020b; Rong et al., 2019). Despite these advancements, there is still no universal solution for constructing very deep GNN like CNN, and relevant works propose different strategies. Lai et al. (Lai et al., 2020)

develops a meta-policy to adaptively select the number of propagation for each graph node through training with reinforcement learning (RL). The experiment results show that partial nodes require more than three layers of propagation to boost model performance. Rong

et al. (Rong et al., 2019) alleviates the over-smoothing problem by randomly removing graph edges, which acts as a message-passing reducer. Claudio et al. (Gallicchio and Micheli, 2020) considers GNN as a dynamic system, and learned representations are the system’s fixed points. Following this assumption, the transformation matrix in propagation is first fixed under stability conditions. Furthermore, only embeddings are updated in the learning procedure, leaving the matrix untrained. In this way, GNN can be trained faster as well as go deeper. Li et al. (Li et al., 2019c) transfer the concepts of residual/dense connections and dilated convolutions from CNNs to assist deeper GNNs. As for future works, the performance leap compared with current shallow GNNs should be an essential problem in developing very deep GNNs, like groundbreaking works in the area of CNN (Szegedy et al., 2015; Huang et al., 2017). At the same time, the computation and time complexity must also be acceptable.

5.2. Dynamic GNN in Recommendation

Existing GNN-based recommendation models are almost based on the static graph, as mentioned above, while there is plenty of dynamics in recommender systems. For example, in the sequential recommendation or session-based recommendation, the users’ data is collected in a dynamic manner, naturally. In addition, modeling the dynamic user preferences is one of the most critical challenges in these recommendation scenarios. In addition, the platform may dynamically involve new users, new products, new features, etc., which poses challenges to static graph neural networks. Recently, dynamic graph neural networks (Li et al., 2020a; Ma et al., 2020a) have attracted attention, which deploys graph neural networks-based models on dynamically constructed graphs. Given the time-evolving property of recommender systems, the dynamic graph neural network-based recommendation model will be a promising research direction, with broad applications in the real world.

5.3. KG-enhanced recommendation with GNN

Recent research has demonstrated the power of KG in recommendation by enriching the user-item bipartite graph with knowledge  (Gao et al., 2020). Specifically, the utilization of knowledge graph with GNN significantly addresses some practical issues in recommender systems (Gao et al., 2020), for example, cold start problem (Fan et al., 2019b; Zhang et al., 2019a), scalability (Xu et al., 2019b), and dynamicity (Sun et al., 2018b). Moreover, knowledge graph also offers a novel solution to some scenarios in recommendations, e.g., sequential recommendation (Huang et al., 2019; Wang et al., 2019c), and objectives, e.g., explainability (Yang and Dong, 2020; Wang et al., 2019a). Currently, the majority of research first uses GNN to learn embeddings of knowledge graphs and then incorporate these embeddings into the recommendation model, so that an end-to-end model can be trained (Wang and Cai, 2020). Therefore, we point out the utilization of KG in recommendation with GNN can be further enhanced from perspectives of data, scenario, and model. Specifically, the incorporated KG mostly records the rich item-item relations, e.g., Movie A belongs to Category B (Gao et al., 2020), but user-user knowledge is lacking in formal and plausible definition and thus substantially overlooked. Future work can consider creating user-centric KG on the foundation of abundant knowledge in sociology. Moreover, considering its success in scenarios, such as sequential recommendation (Huang et al., 2019; Wang et al., 2019c), it is promising that the leverage of KG could further enhance recommendation quality from the aspects that require more external knowledge, such as diversity and fairness. Further, existing methods on leveraging KG in recommendation cannot fully model complex relations between a user and a specific item or its attributes. Thus, designing a better framework to carve out these complex relations is another future direction.

5.4. Efficiency and Scalability

Early works on GNN follow the full-batch gradient descent algorithm, where the whole adjacency matrix is multiplied on the node embeddings during each inference step, which can not handle real-world recommender systems since the number of nodes and edges can reach a million-level scale. Hamilton et al. propose GraphSAGE (Hamilton et al., 2017) which performs neighbor sampling and only updates the related sub-graph instead of the whole graph during each inference step. The sampling strategy is also adopted in a few other works (Chiang et al., 2019; Chen et al., 2018) which reduce the computation complexity of GNN and improve the scalability. Ying et al. (Ying et al., 2018)

successfully apply GraphSAGE to web-scale recommender systems, which can efficiently compute embeddings for billions of items. In addition, a few open-source tools have been released which can accelerate the research and development of GNN-based recommendation, such as PyG

(Fey and Lenssen, 2019), DGL (Wang et al., 2019d) and AliGraph (Zhu et al., 2019). We refer to another survey (Abadal et al., 2020) for details on computing and accelerating GNN. Despite these existing approaches, how to achieve large-scale GNN based recommendation is still a challenging task, especially in the ranking stage where thousands of features are involved, which results in a large and complicated heterogeneous graph.

5.5. Self-supervised GNN

The direct supervision from interaction data is relatively sparse compared with the scale of the graph. Therefore, it is necessary to include more supervision signals from the graph structure itself or the recommendation task. For example, Yu et al. (Yu et al., 2021c) and Wu et al. (Wu et al., 2021c) attempt to enhance GNN-based recommendation by designing auxiliary tasks from the graph structure with self-supervision. Data augmentations such as node dropout are utilized to generate sample pairs for contrastive training. We believe that it is a promising future direction to leverage extra self-supervised tasks to learn meaningful and robust representations for GNN-based recommender systems.

5.6. Conversational Recommendation with GNN

In existing recommender systems, there may exist the issue of information asymmetry that the system can only estimate users’ preferences based on their historically collected behavior data. To address it, recently, conversational (interactive) recommendation researches (Sun and Zhang, 2018; Lei et al., 2020), propose the new paradigm the user can interact with the system in conversations, and then new data can be dynamically collected. Specifically, users can chat with the system to explicitly convey their consumption demands or offer positive/negative feedback on the recommended items. As future work, the advances of representation learning with graph neural networks can be combined with preference learning in the conversational recommendation.

5.7. AutoML-enhanced GNN

Recommendation scenarios are diverse and largely different from each other; thus, there exists no silver bullet GNN model that can generalize across all scenarios. Recently, AutoML (Automated Machine Learning) (Yao et al., 2018) is proposed, which can automatically design appropriate models for specific tasks. With respect to GNN-based recommendation, the search space is quite large, which includes multiple options for the neighbor sampler, aggregator, interaction function, and so on. As a consequence, AutoML can, to a great extent, reduce human effort in discovering advanced model structures. A few works (Gao et al., 2019b; Huan et al., 2021) have been proposed which search to combine GNN layers and aggregate neighbors. However, designing AutoML algorithms to search for GNN based recommender systems is a largely unexplored yet promising future direction.

6. Conclusion

There is a rapid development of graph neural networks models in the research field of recommender systems. This paper provides an extensive survey systematically presenting the challenges, methods, and future directions in this area. Not only the history of development and also the most recent advances are well covered and introduced. We hope this survey can well help both junior and experienced researchers in the relative areas.

References

  • (1)
  • Abadal et al. (2020) Sergi Abadal, Akshay Jain, Robert Guirado, Jorge López-Alonso, and Eduard Alarcón. 2020. Computing graph neural networks: A survey from algorithms to accelerators. arXiv preprint arXiv:2010.00130 (2020).
  • Abdollahpouri et al. (2017) Himan Abdollahpouri, Robin Burke, and Bamshad Mobasher. 2017. Controlling popularity bias in learning-to-rank recommendation. In Proceedings of the eleventh ACM conference on recommender systems. 42–46.
  • Abdollahpouri et al. (2019) Himan Abdollahpouri, Masoud Mansoury, Robin Burke, and Bamshad Mobasher. 2019. The unfairness of popularity bias in recommendation. arXiv preprint arXiv:1907.13286 (2019).
  • Abdollahpouri et al. (2020) Himan Abdollahpouri, Masoud Mansoury, Robin Burke, and Bamshad Mobasher. 2020. The Connection Between Popularity Bias, Calibration, and Fairness in Recommendation. In Fourteenth ACM Conference on Recommender Systems. Association for Computing Machinery, New York, NY, USA, 726–731.
  • Ai et al. (2018) Qingyao Ai, Vahid Azizi, Xu Chen, and Yongfeng Zhang. 2018. Learning heterogeneous knowledge base embeddings for explainable recommendation. Algorithms 11, 9 (2018), 137.
  • Bai et al. (2020) Ting Bai, Youjie Zhang, Bin Wu, and Jian-Yun Nie. 2020. Temporal Graph Neural Networks for Social Recommendation. In 2020 IEEE International Conference on Big Data (Big Data). IEEE, 898–903.
  • Bell and Koren (2007) Robert M Bell and Yehuda Koren. 2007. Lessons from the netflix prize challenge. Acm Sigkdd Explorations Newsletter 9, 2 (2007), 75–79.
  • Berg et al. (2018) Rianne van den Berg, Thomas N Kipf, and Max Welling. 2018. Graph convolutional matrix completion. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining.
  • Beutel et al. (2019) Alex Beutel, Jilin Chen, Tulsee Doshi, Hai Qian, Li Wei, Yi Wu, Lukasz Heldt, Zhe Zhao, Lichan Hong, Ed H Chi, et al. 2019. Fairness in Recommendation Ranking through Pairwise Comparisons. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. Association for Computing Machinery, New York, NY, USA, 2212–2220.
  • Bose and Hamilton (2019) Avishek Bose and William Hamilton. 2019. Compositional fairness constraints for graph embeddings. In International Conference on Machine Learning. PMLR, 715–724.
  • Cao et al. (2017) Da Cao, Liqiang Nie, Xiangnan He, Xiaochi Wei, Shunzhi Zhu, and Tat-Seng Chua. 2017. Embedding factorization models for jointly recommending items and user generated lists. In SIGIR. 585–594.
  • Chang et al. (2020b) Buru Chang, Gwanghoon Jang, Seoyoon Kim, and Jaewoo Kang. 2020b. Learning graph-based geographical latent representation for point-of-interest recommendation. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 135–144.
  • Chang et al. (2020a) Jianxin Chang, Chen Gao, Xiangnan He, Depeng Jin, and Yong Li. 2020a. Bundle recommendation with graph convolutional networks. In Proceedings of the 43rd international ACM SIGIR conference on Research and development in Information Retrieval. 1673–1676.
  • Chang et al. (2021) Jianxin Chang, Chen Gao, Yu Zheng, Yiqun Hui, Yanan Niu, Yang Song, Depeng Jin, and Yong Li. 2021. Sequential Recommendation with Graph Neural Networks. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 378–387.
  • Chen et al. (2021a) Chen Chen, Jie Guo, and Bin Song. 2021a. Dual Attention Transfer in Session-based Recommendation with Multi-dimensional Integration. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 869–878.
  • Chen et al. (2021c) Chong Chen, Weizhi Ma, Min Zhang, Zhaowei Wang, Xiuqiang He, Chenyang Wang, Yiqun Liu, and Shaoping Ma. 2021c. Graph Heterogeneous Multi-Relational Recommendation. In

    Proceedings of the AAAI Conference on Artificial Intelligence

    , Vol. 35. 3958–3966.
  • Chen et al. (2019b) Chong Chen, Min Zhang, Chenyang Wang, Weizhi Ma, Minming Li, Yiqun Liu, and Shaoping Ma. 2019b. An efficient adaptive transfer neural network for social-aware recommendation. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. 225–234.
  • Chen et al. (2020a) Deli Chen, Yankai Lin, Wei Li, Peng Li, Jie Zhou, and Xu Sun. 2020a. Measuring and relieving the over-smoothing problem for graph neural networks from the topological view. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 34. 3438–3445.
  • Chen et al. (2020b) Deli Chen, Yankai Lin, Wei Li, Peng Li, Jie Zhou, and Xu Sun. 2020b. Measuring and relieving the over-smoothing problem for graph neural networks from the topological view. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 34. 3438–3445.
  • Chen et al. (2021b) Hongxu Chen, Yicong Li, Xiangguo Sun, Guandong Xu, and Hongzhi Yin. 2021b. Temporal Meta-Path Guided Explainable Recommendation. In Proceedings of the 14th ACM International Conference on Web Search and Data Mining (WSDM ’21). Association for Computing Machinery, New York, NY, USA, 1056–1064.
  • Chen et al. (2018) Jie Chen, Tengfei Ma, and Cao Xiao. 2018. Fastgcn: fast learning with graph convolutional networks via importance sampling. arXiv preprint arXiv:1801.10247 (2018).
  • Chen et al. (2019a) Liang Chen, Yang Liu, Xiangnan He, Lianli Gao, and Zibin Zheng. 2019a. Matching user with item set: collaborative bundle recommendation with deep attention network. In IJCAI. 2095–2101.
  • Chen and Wong (2020) Tianwen Chen and Raymond Chi-Wing Wong. 2020. Handling information loss of graph neural networks for session-based recommendation. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1172–1180.
  • Chen and Wong (2021) Tianwen Chen and Raymond Chi-Wing Wong. 2021. An Efficient and Effective Framework for Session-based Social Recommendation. In Proceedings of the 14th ACM International Conference on Web Search and Data Mining. 400–408.
  • Cheng et al. (2013) Chen Cheng, Haiqin Yang, Michael R Lyu, and Irwin King. 2013. Where you like to go next: Successive point-of-interest recommendation. In Twenty-Third international joint conference on Artificial Intelligence.
  • Cheng et al. (2016) Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. 2016. Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems. 7–10.
  • Chiang et al. (2019) Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui Hsieh. 2019. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 257–266.
  • Cialdini and Goldstein (2004) Robert B Cialdini and Noah J Goldstein. 2004. Social influence: Compliance and conformity. Annu. Rev. Psychol. 55 (2004), 591–621.
  • Covington et al. (2016) Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems. 191–198.
  • Cui et al. (2020) Qiang Cui, Tao Wei, Yafeng Zhang, and Qing Zhang. 2020. HeroGRAPH: A Heterogeneous Graph Framework for Multi-Target Cross-Domain Recommendation.. In ORSUM@ RecSys.
  • Dai and Wang (2021) Enyan Dai and Suhang Wang. 2021. Say No to the Discrimination: Learning Fair Graph Neural Networks with Limited Sensitive Attribute Information (WSDM ’21). Association for Computing Machinery, New York, NY, USA, 680–688.
  • Davidson et al. (2010) James Davidson, Benjamin Liebald, Junning Liu, Palash Nandy, Taylor Van Vleet, Ullas Gargi, Sujoy Gupta, Yu He, Mike Lambert, Blake Livingston, et al. 2010. The YouTube video recommendation system. In Proceedings of the fourth ACM conference on Recommender systems. 293–296.
  • Deng et al. (2020) Qilin Deng, Kai Wang, Minghao Zhao, Zhene Zou, Runze Wu, Jianrong Tao, Changjie Fan, and Liang Chen. 2020. Personalized Bundle Recommendation in Online Games. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 2381–2388.
  • Fan et al. (2019b) Shaohua Fan, Junxiong Zhu, Xiaotian Han, Chuan Shi, Linmei Hu, Biyu Ma, and Yongliang Li. 2019b. Metapath-Guided Heterogeneous Graph Neural Network for Intent Recommendation. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery &; Data Mining (KDD ’19). Association for Computing Machinery, New York, NY, USA, 2478–2486.
  • Fan et al. (2019a) Wenqi Fan, Yao Ma, Qing Li, Yuan He, Eric Zhao, Jiliang Tang, and Dawei Yin. 2019a. Graph neural networks for social recommendation. In The World Wide Web Conference. 417–426.
  • Fan et al. (2021) Ziwei Fan, Zhiwei Liu, Jiawei Zhang, Yun Xiong, Lei Zheng, and Philip S Yu. 2021. Continuous-Time Sequential Recommendation with Temporal Graph Collaborative Transformer. arXiv preprint arXiv:2108.06625 (2021).
  • Feng et al. (2015) Shanshan Feng, Xutao Li, Yifeng Zeng, Gao Cong, Yeow Meng Chee, and Quan Yuan. 2015. Personalized ranking metric embedding for next new poi recommendation. In Twenty-Fourth International Joint Conference on Artificial Intelligence.
  • Feng et al. (2019) Yifan Feng, Haoxuan You, Zizhao Zhang, Rongrong Ji, and Yue Gao. 2019. Hypergraph neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 3558–3565.
  • Fey and Lenssen (2019) Matthias Fey and Jan Eric Lenssen. 2019. Fast graph representation learning with PyTorch Geometric. arXiv preprint arXiv:1903.02428 (2019).
  • Fout (2017) Alex M Fout. 2017. Protein interface prediction using graph convolutional networks. Ph.D. Dissertation. Colorado State University.
  • Fu et al. (2019) Wenjing Fu, Zhaohui Peng, Senzhang Wang, Yang Xu, and Jin Li. 2019. Deeply fusing reviews and contents for cold start users in cross-domain recommendation systems. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 94–101.
  • Gallicchio and Micheli (2020) Claudio Gallicchio and Alessio Micheli. 2020. Fast and deep graph neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 34. 3898–3905.
  • Gao et al. (2019a) Chen Gao, Xiangning Chen, Fuli Feng, Kai Zhao, Xiangnan He, Yong Li, and Depeng Jin. 2019a. Cross-domain recommendation without sharing user-relevant data. In The world wide web conference. 491–502.
  • Gao et al. (2020) Yang Gao, Yi-Fan Li, Yu Lin, Hang Gao, and Latifur Khan. 2020. Deep learning on knowledge graph for recommender system: A survey. arXiv preprint arXiv:2004.00387 (2020).
  • Gao et al. (2019b) Yang Gao, Hong Yang, Peng Zhang, Chuan Zhou, and Yue Hu. 2019b. Graphnas: Graph neural architecture search with reinforcement learning. arXiv preprint arXiv:1904.09981 (2019).
  • Geyik et al. (2019) Sahin Cem Geyik, Stuart Ambler, and Krishnaram Kenthapadi. 2019. Fairness-Aware Ranking in Search & Recommendation Systems with Application to LinkedIn Talent Search. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD ’19). Association for Computing Machinery, New York, NY, USA, 2221–2231.
  • Grover and Leskovec (2016) Aditya Grover and Jure Leskovec. 2016. node2vec: Scalable feature learning for networks. In Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining. 855–864.
  • Guo et al. (2017) Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: a factorization-machine based neural network for CTR prediction. arXiv preprint arXiv:1703.04247 (2017).
  • Guo et al. (2021b) Lei Guo, Li Tang, Tong Chen, Lei Zhu, Quoc Viet Hung Nguyen, and Hongzhi Yin. 2021b. DA-GCN: A Domain-aware Attentive Graph Convolution Network for Shared-account Cross-domain Sequential Recommendation. arXiv preprint arXiv:2105.03300 (2021).
  • Guo et al. (2021a) Wei Guo, Rong Su, Renhao Tan, Huifeng Guo, Yingxue Zhang, Zhirong Liu, Ruiming Tang, and Xiuqiang He. 2021a. Dual Graph enhanced Embedding Neural Network for CTRPrediction. arXiv preprint arXiv:2106.00314 (2021).
  • Guo and Wang (2020) Zhiwei Guo and Heng Wang. 2020. A deep graph neural network-based mechanism for social recommendations. IEEE Transactions on Industrial Informatics 17, 4 (2020), 2776–2783.
  • Hamilton et al. (2017) William L Hamilton, Rex Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. In Proceedings of the 31st International Conference on Neural Information Processing Systems. 1025–1035.
  • Hammond et al. (2011) David K Hammond, Pierre Vandergheynst, and Rémi Gribonval. 2011. Wavelets on graphs via spectral graph theory. Applied and Computational Harmonic Analysis 30, 2 (2011), 129–150.
  • He et al. (2015) Xiangnan He, Tao Chen, Min-Yen Kan, and Xiao Chen. 2015. Trirank: Review-aware explainable recommendation by modeling aspects. In Proceedings of the 24th ACM International on Conference on Information and Knowledge Management. 1661–1670.
  • He et al. (2020) Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. In Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 639–648.
  • He et al. (2017) Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. In Proceedings of the 26th international conference on world wide web. 173–182.
  • Hidasi et al. (2015) Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk. 2015. Session-based recommendations with recurrent neural networks. arXiv preprint arXiv:1511.06939 (2015).
  • Hochreiter and Schmidhuber (1997) Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation 9, 8 (1997), 1735–1780.
  • Hsu and Li (2021) Cheng Hsu and Cheng-Te Li. 2021. RetaGNN: Relational Temporal Attentive Graph Neural Networks for Holistic Sequential Recommendation. In Proceedings of the Web Conference 2021. 2968–2979.
  • Hu et al. (2018) Guangneng Hu, Yu Zhang, and Qiang Yang. 2018. Conet: Collaborative cross networks for cross-domain recommendation. In Proceedings of the 27th ACM international conference on information and knowledge management. 667–676.
  • Hu et al. (2020) Linmei Hu, Siyong Xu, Chen Li, Cheng Yang, Chuan Shi, Nan Duan, Xing Xie, and Ming Zhou. 2020. Graph neural news recommendation with unsupervised preference disentanglement. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 4255–4264.
  • Huan et al. (2021) ZHAO Huan, YAO Quanming, and TU Weiwei. 2021. Search to aggregate neighborhood for graph neural network. In 2021 IEEE 37th International Conference on Data Engineering (ICDE). IEEE, 552–563.
  • Huang et al. (2021) Chao Huang, Huance Xu, Yong Xu, Peng Dai, Lianghao Xia, Mengyin Lu, Liefeng Bo, Hao Xing, Xiaoping Lai, and Yanfang Ye. 2021. Knowledge-aware coupled graph neural network for social recommendation. In AAAI Conference on Artificial Intelligence (AAAI).
  • Huang et al. (2017) Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. 2017. Densely connected convolutional networks. In

    Proceedings of the IEEE conference on computer vision and pattern recognition

    . 4700–4708.
  • Huang et al. (2019) Xiaowen Huang, Quan Fang, Shengsheng Qian, Jitao Sang, Yan Li, and Changsheng Xu. 2019. Explainable Interaction-Driven User Modeling over Knowledge Graph for Sequential Recommendation. In Proceedings of the 27th ACM International Conference on Multimedia (MM ’19). Association for Computing Machinery, New York, NY, USA, 548–556.
  • Isufi et al. (2021) Elvin Isufi, Matteo Pocchiari, and Alan Hanjalic. 2021. Accuracy-diversity trade-off in recommender systems via graph convolutions. Information Processing & Management 58, 2 (2021), 102459.
  • Jin et al. (2020) Bowen Jin, Chen Gao, Xiangnan He, Depeng Jin, and Yong Li. 2020. Multi-behavior recommendation with graph convolutional networks. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 659–668.
  • Kang et al. (2019) SeongKu Kang, Junyoung Hwang, Dongha Lee, and Hwanjo Yu. 2019. Semi-supervised learning for cross-domain recommendation to cold-start users. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 1563–1572.
  • Kang and McAuley (2018) Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recommendation. In 2018 IEEE International Conference on Data Mining (ICDM). IEEE, 197–206.
  • Kang and McAuley (2019) Wang-Cheng Kang and Julian McAuley. 2019. Candidate generation with binary codes for large-scale top-n recommendation. In Proceedings of the 28th ACM international conference on information and knowledge management. 1523–1532.
  • Kim et al. (2019) Kyung-Min Kim, Donghyun Kwak, Hanock Kwak, Young-Jin Park, Sangkwon Sim, Jae-Han Cho, Minkyu Kim, Jihun Kwon, Nako Sung, and Jung-Woo Ha. 2019. Tripartite heterogeneous graph propagation for large-scale social recommendation. arXiv preprint arXiv:1908.02569 (2019).
  • Kipf and Welling (2017) Thomas N Kipf and Max Welling. 2017. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations.
  • Koren (2009) Yehuda Koren. 2009. The bellkor solution to the netflix grand prize. Netflix prize documentation 81, 2009 (2009), 1–10.
  • Koren et al. (2009) Yehuda Koren, Robert Bell, and Chris Volinsky. 2009. Matrix factorization techniques for recommender systems. Computer 42, 8 (2009), 30–37.
  • Kunkel et al. (2019) Johannes Kunkel, Tim Donkers, Lisa Michael, Catalin-Mihai Barbu, and Jürgen Ziegler. 2019. Let me explain: impact of personal and impersonal explanations on trust in recommender systems. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery, New York, NY, USA, 1–12.
  • Lai et al. (2020) Kwei-Herng Lai, Daochen Zha, Kaixiong Zhou, and Xia Hu. 2020. Policy-gnn: Aggregation optimization for graph neural networks. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 461–471.
  • Lei et al. (2020) Wenqiang Lei, Gangyi Zhang, Xiangnan He, Yisong Miao, Xiang Wang, Liang Chen, and Tat-Seng Chua. 2020. Interactive path reasoning on graph for conversational recommendation. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2073–2083.
  • Leonhardt et al. (2018) Jurek Leonhardt, Avishek Anand, and Megha Khosla. 2018. User fairness in recommender systems. In Companion Proceedings of the The Web Conference 2018. 101–102.
  • Li et al. (2021d) Chen Li, Linmei Hu, Chuan Shi, Guojie Song, and Yuanfu Lu. 2021d. Sequence-aware Heterogeneous Graph Neural Collaborative Filtering. In Proceedings of the 2021 SIAM International Conference on Data Mining (SDM). SIAM, 64–72.
  • Li et al. (2019c) Guohao Li, Matthias Muller, Ali Thabet, and Bernard Ghanem. 2019c. Deepgcns: Can gcns go as deep as cnns?. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 9267–9276.
  • Li et al. (2020a) Maosen Li, Siheng Chen, Yangheng Zhao, Ya Zhang, Yanfeng Wang, and Qi Tian. 2020a. Dynamic multiscale graph neural networks for 3d skeleton based human motion prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 214–223.
  • Li et al. (2019b) Mengmeng Li, Tian Gan, Meng Liu, Zhiyong Cheng, Jianhua Yin, and Liqiang Nie. 2019b. Long-tail hashtag recommendation for micro-videos with graph convolutional network. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 509–518.
  • Li et al. (2020c) Xingchen Li, Xiang Wang, Xiangnan He, Long Chen, Jun Xiao, and Tat-Seng Chua. 2020c. Hierarchical fashion graph network for personalized outfit recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 159–168.
  • Li et al. (2021a) Yunqi Li, Hanxiong Chen, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2021a. User-oriented Fairness in Recommendation. In Proceedings of the Web Conference 2021. 624–632.
  • Li et al. (2021b) Yang Li, Tong Chen, Hongzhi Yin, and Zi Huang. 2021b. Discovering Collaborative Signals for Next POI Recommendation with Iterative Seq2Graph Augmentation. arXiv preprint arXiv:2106.15814 (2021).
  • Li et al. (2021c) Yunqi Li, Yingqiang Ge, and Yongfeng Zhang. 2021c. Tutorial on Fairness of Machine Learning in Recommender Systems. SIGIR.
  • Li et al. (2019a) Zekun Li, Zeyu Cui, Shu Wu, Xiaoyu Zhang, and Liang Wang. 2019a. Fi-gnn: Modeling feature interactions via graph neural networks for ctr prediction. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 539–548.
  • Li et al. (2020b) Zhao Li, Xin Shen, Yuhang Jiao, Xuming Pan, Pengcheng Zou, Xianling Meng, Chengwei Yao, and Jiajun Bu. 2020b. Hierarchical bipartite graph neural networks: Towards large-scale e-commerce applications. In 2020 IEEE 36th International Conference on Data Engineering (ICDE). IEEE, 1677–1688.
  • Lian et al. (2018) Jianxun Lian, Xiaohuan Zhou, Fuzheng Zhang, Zhongxia Chen, Xing Xie, and Guangzhong Sun. 2018. xdeepfm: Combining explicit and implicit feature interactions for recommender systems. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1754–1763.
  • Lim et al. (2020) Nicholas Lim, Bryan Hooi, See-Kiong Ng, Xueou Wang, Yong Liang Goh, Renrong Weng, and Jagannadan Varadarajan. 2020. STP-UDGAT: Spatial-Temporal-Preference User Dimensional Graph Attention Network for Next POI Recommendation. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 845–854.
  • Liu et al. (2020c) Feng Liu, Weiwen Liu, Xutao Li, and Yunming Ye. 2020c. Inter-sequence Enhanced Framework for Personalized Sequential Recommendation. arXiv preprint arXiv:2004.12118 (2020).
  • Liu et al. (2020a) Meng Liu, Jianjun Li, Guohui Li, and Peng Pan. 2020a. Cross Domain Recommendation via Bi-directional Transfer Graph Collaborative Filtering Networks. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 885–894.
  • Liu et al. (2020e) Qi Liu, Ruobing Xie, Lei Chen, Shukai Liu, Ke Tu, Peng Cui, Bo Zhang, and Leyu Lin. 2020e. Graph neural network for tag ranking in tag-enhanced video recommendation. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 2613–2620.
  • Liu et al. (2020d) Weiwen Liu, Qing Liu, Ruiming Tang, Junyang Chen, Xiuqiang He, and Pheng Ann Heng. 2020d. Personalized Re-ranking with Item Relationships for E-commerce. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 925–934.
  • Liu et al. (2021) Weiwen Liu, Yin Zhang, Jianling Wang, Yun He, James Caverlee, Patrick PK Chan, Daniel S Yeung, and Pheng-Ann Heng. 2021. Item Relationship Graph Neural Networks for E-Commerce. IEEE Transactions on Neural Networks and Learning Systems (2021).
  • Liu et al. (2020b) Yang Liu, Chen Liang, Xiangnan He, Jiaying Peng, Zibin Zheng, and Jie Tang. 2020b. Modelling high-order social relations for item recommendation. IEEE Transactions on Knowledge and Data Engineering (2020).
  • Liu et al. (2014) Yidan Liu, Min Xie, and Laks VS Lakshmanan. 2014. Recommending user generated item lists. In RecSys. 185–192.
  • Luo et al. (2020a) Dongsheng Luo, Yuchen Bian, Xiang Zhang, and Jun Huan. 2020a. Attentive Social Recommendation: Towards User And Item Diversities. arXiv preprint arXiv:2011.04797 (2020).
  • Luo et al. (2020b) Xusheng Luo, Luxin Liu, Yonghua Yang, Le Bo, Yuanpeng Cao, Jinghang Wu, Qiang Li, Keping Yang, and Kenny Q Zhu. 2020b. AliCoCo: Alibaba e-commerce cognitive concept net. In Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data. 313–327.
  • Ma et al. (2020b) Chen Ma, Liheng Ma, Yingxue Zhang, Jianing Sun, Xue Liu, and Mark Coates. 2020b. Memory augmented graph neural networks for sequential recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 34. 5045–5052.
  • Ma et al. (2019) Weizhi Ma, Min Zhang, Yue Cao, Woojeong Jin, Chenyang Wang, Yiqun Liu, Shaoping Ma, and Xiang Ren. 2019. Jointly Learning Explainable Rules for Recommendation with Knowledge Graph. In The World Wide Web Conference (WWW ’19). Association for Computing Machinery, New York, NY, USA, 1210–1221.
  • Ma et al. (2018) Xiao Ma, Liqin Zhao, Guan Huang, Zhi Wang, Zelin Hu, Xiaoqiang Zhu, and Kun Gai. 2018. Entire space multi-task model: An effective approach for estimating post-click conversion rate. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 1137–1140.
  • Ma et al. (2020a) Yao Ma, Ziyi Guo, Zhaocun Ren, Jiliang Tang, and Dawei Yin. 2020a. Streaming graph neural networks. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 719–728.
  • Man et al. (2017) Tong Man, Huawei Shen, Xiaolong Jin, and Xueqi Cheng. 2017. Cross-Domain Recommendation: An Embedding and Mapping Approach.. In IJCAI, Vol. 17. 2464–2470.
  • Mansoury et al. (2020) Masoud Mansoury, Himan Abdollahpouri, Mykola Pechenizkiy, Bamshad Mobasher, and Robin Burke. 2020. FairMatch: A Graph-Based Approach for Improving Aggregate Diversity in Recommender Systems. In Proceedings of the 28th ACM Conference on User Modeling, Adaptation and Personalization (UMAP ’20). Association for Computing Machinery, New York, NY, USA, 154–162. https://doi.org/10.1145/3340631.3394860
  • McAuley and Leskovec (2013) Julian McAuley and Jure Leskovec. 2013. Hidden factors and hidden topics: understanding rating dimensions with review text. In Proceedings of the 7th ACM conference on Recommender systems. 165–172.
  • McPherson et al. (2001) Miller McPherson, Lynn Smith-Lovin, and James M Cook. 2001. Birds of a feather: Homophily in social networks. Annual review of sociology 27, 1 (2001), 415–444.
  • Mehrotra et al. (2018) Rishabh Mehrotra, James McInerney, Hugues Bouchard, Mounia Lalmas, and Fernando Diaz. 2018. Towards a Fair Marketplace: Counterfactual Evaluation of the Trade-off between Relevance, Fairness & Satisfaction in Recommendation Systems. In Proceedings of the 27th ACM International Conference on Information and Knowledge Management. Association for Computing Machinery, New York, NY, USA, 2243–2251.
  • Meng et al. (2020) Wenjing Meng, Deqing Yang, and Yanghua Xiao. 2020. Incorporating user micro-behaviors and item knowledge into multi-task learning for session-based recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 1091–1100.
  • Mikolov et al. (2013) Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. In International Conference on Learning Representations.
  • Min et al. (2019) Weiqing Min, Shuqiang Jiang, and Ramesh Jain. 2019. Food Recommendation: Framework, Existing Solutions, and Challenges. IEEE Transactions on Multimedia 22, 10 (2019), 2659–2671.
  • Mirbakhsh and Ling (2015) Nima Mirbakhsh and Charles X Ling. 2015. Improving top-n recommendation for cold-start users via cross-domain information. ACM Transactions on Knowledge Discovery from Data (TKDD) 9, 4 (2015), 1–19.
  • Mu et al. (2019) Nan Mu, Daren Zha, Yuanye He, and Zhihao Tang. 2019. Graph attention networks for neural social recommendation. In 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI). IEEE, 1320–1327.
  • Okura et al. (2017) Shumpei Okura, Yukihiro Tagami, Shingo Ono, and Akira Tajima. 2017. Embedding-based news recommendation for millions of users. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 1933–1942.
  • Pal et al. (2019) Soumyasundar Pal, Florence Regol, and Mark Coates. 2019. Bayesian graph convolutional neural networks using node copying. arXiv preprint arXiv:1911.04965 (2019).
  • Pan et al. (2020) Zhiqiang Pan, Fei Cai, Wanyu Chen, Honghui Chen, and Maarten de Rijke. 2020. Star graph neural networks for session-based recommendation. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 1195–1204.
  • Paparrizos et al. (2011) Ioannis Paparrizos, B Barla Cambazoglu, and Aristides Gionis. 2011. Machine learned job recommendation. In Proceedings of the fifth ACM Conference on Recommender Systems. 325–328.
  • Pathak et al. (2017) Apurva Pathak, Kshitiz Gupta, and Julian McAuley. 2017. Generating and personalizing bundle recommendations on steam. In SIGIR. 1073–1076.
  • Peake and Wang (2018) Georgina Peake and Jun Wang. 2018. Explanation mining: Post hoc interpretability of latent factor models for recommendation systems. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2060–2069.
  • Pei et al. (2019) Changhua Pei, Yi Zhang, Yongfeng Zhang, Fei Sun, Xiao Lin, Hanxiao Sun, Jian Wu, Peng Jiang, Junfeng Ge, Wenwu Ou, et al. 2019. Personalized re-ranking for recommendation. In Proceedings of the 13th ACM Conference on Recommender Systems. 3–11.
  • Perozzi et al. (2014) Bryan Perozzi, Rami Al-Rfou, and Steven Skiena. 2014. Deepwalk: Online learning of social representations. In Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining. 701–710.
  • Qiu et al. (2019) Ruihong Qiu, Jingjing Li, Zi Huang, and Hongzhi Yin. 2019. Rethinking the item order in session-based recommendation with graph neural networks. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 579–588.
  • Qiu et al. (2020) Ruihong Qiu, Hongzhi Yin, Zi Huang, and Tong Chen. 2020. Gag: Global attributed graph neural network for streaming session-based recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 669–678.
  • Rahman et al. (2019) Tahleen Rahman, Bartlomiej Surma, Michael Backes, and Yang Zhang. 2019. Fairwalk: Towards Fair Graph Embedding. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19. International Joint Conferences on Artificial Intelligence Organization, 3289–3295. https://doi.org/10.24963/ijcai.2019/456
  • Rendle (2010) Steffen Rendle. 2010. Factorization machines. In 2010 IEEE International conference on data mining. IEEE, 995–1000.
  • Rendle et al. (2009) Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. 2009. BPR: Bayesian personalized ranking from implicit feedback. In Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence. 452–461.
  • Rong et al. (2019) Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. 2019. DropEdge: Towards Deep Graph Convolutional Networks on Node Classification. In International Conference on Learning Representations.
  • Sheu and Li (2020) Heng-Shiou Sheu and Sheng Li. 2020. Context-aware graph embedding for session-based news recommendation. In Fourteenth ACM conference on recommender systems. 657–662.
  • Shuman et al. (2013) David I Shuman, Sunil K Narang, Pascal Frossard, Antonio Ortega, and Pierre Vandergheynst. 2013.

    The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains.

    IEEE signal processing magazine 30, 3 (2013), 83–98.
  • Singh and Joachims (2018) Ashudeep Singh and Thorsten Joachims. 2018. Fairness of Exposure in Rankings. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD ’18). Association for Computing Machinery, New York, NY, USA, 2219–2228.
  • Sinha and Swearingen (2002) Rashmi Sinha and Kirsten Swearingen. 2002. The role of transparency in recommender systems. In CHI’02 extended abstracts on Human factors in computing systems. Association for Computing Machinery, New York, NY, USA, 830–831.
  • Song et al. (2021) Changhao Song, Bo Wang, Qinxue Jiang, Yehua Zhang, Ruifang He, and Yuexian Hou. 2021. Social Recommendation with Implicit Social Influence. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1788–1792.
  • Song et al. (2019a) Weiping Song, Chence Shi, Zhiping Xiao, Zhijian Duan, Yewen Xu, Ming Zhang, and Jian Tang. 2019a. Autoint: Automatic feature interaction learning via self-attentive neural networks. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 1161–1170.
  • Song et al. (2019b) Weiping Song, Zhiping Xiao, Yifan Wang, Laurent Charlin, Ming Zhang, and Jian Tang. 2019b. Session-based social recommendation via dynamic graph attention networks. In Proceedings of the Twelfth ACM international conference on web search and data mining. 555–563.
  • Su et al. (2021) Yixin Su, Rui Zhang, Sarah Erfani, and Zhenghua Xu. 2021. Detecting Beneficial Feature Interactions for Recommender Systems. In Proceedings of the 34th AAAI Conference on Artificial Intelligence (AAAI).
  • Sun et al. (2020a) Jianing Sun, Wei Guo, Dengcheng Zhang, Yingxue Zhang, Florence Regol, Yaochen Hu, Huifeng Guo, Ruiming Tang, Han Yuan, Xiuqiang He, et al. 2020a. A framework for recommending accurate and diverse items using bayesian graph convolutional neural networks. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2030–2039.
  • Sun et al. (2020b) Jianing Sun, Yingxue Zhang, Wei Guo, Huifeng Guo, Ruiming Tang, Xiuqiang He, Chen Ma, and Mark Coates. 2020b. Neighbor interaction aware graph convolution networks for recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 1289–1298.
  • Sun et al. (2018a) Peijie Sun, Le Wu, and Meng Wang. 2018a. Attentive recurrent social recommendation. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 185–194.
  • Sun and Zhang (2018) Yueming Sun and Yi Zhang. 2018. Conversational recommender system. In The 41st international acm sigir conference on research & development in information retrieval. 235–244.
  • Sun et al. (2018b) Zhu Sun, Jie Yang, Jie Zhang, Alessandro Bozzon, Long-Kai Huang, and Chi Xu. 2018b. Recurrent Knowledge Graph Embedding for Effective Recommendation. In Proceedings of the 12th ACM Conference on Recommender Systems (RecSys ’18). Association for Computing Machinery, New York, NY, USA, 297–305.
  • Szegedy et al. (2015) Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. 2015. Going deeper with convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition. 1–9.
  • Tintarev and Masthoff (2007) Nava Tintarev and Judith Masthoff. 2007. A survey of explanations in recommender systems. In 2007 IEEE 23rd international conference on data engineering workshop. IEEE, 801–810.
  • Van Den Oord et al. (2013) Aäron Van Den Oord, Sander Dieleman, and Benjamin Schrauwen. 2013. Deep content-based music recommendation. In Neural Information Processing Systems Conference (NIPS 2013), Vol. 26. Neural Information Processing Systems Foundation (NIPS).
  • Veličković et al. (2018) Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. 2018. Graph attention networks. In International Conference on Learning Representations.
  • Wang and Cai (2020) Baocheng Wang and Wentao Cai. 2020. Knowledge-enhanced graph neural networks for sequential recommendation. Information 11, 8 (2020), 388.
  • Wang et al. (2020d) Bei Wang, Chenrui Zhang, Hao Zhang, Xiaoqing Lyu, and Zhi Tang. 2020d.

    Dual Autoencoder Network with Swap Reconstruction for Cold-Start Recommendation. In

    Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 2249–2252.
  • Wang et al. (2018) Hongwei Wang, Fuzheng Zhang, Jialin Wang, Miao Zhao, Wenjie Li, Xing Xie, and Minyi Guo. 2018. RippleNet: Propagating User Preferences on the Knowledge Graph for Recommender Systems. In Proceedings of the 27th ACM International Conference on Information and Knowledge Management (CIKM ’18). Association for Computing Machinery, New York, NY, USA, 417–426.
  • Wang et al. (2021a) Jianling Wang, Kaize Ding, Ziwei Zhu, and James Caverlee. 2021a. Session-based Recommendation with Hypergraph Attention Networks. In Proceedings of the 2021 SIAM International Conference on Data Mining (SDM). SIAM, 82–90.
  • Wang et al. (2019d) Minjie Wang, Lingfan Yu, Da Zheng, Quan Gan, Yu Gai, Zihao Ye, Mufei Li, Jinjing Zhou, Qi Huang, Chao Ma, et al. 2019d. Deep Graph Library: Towards Efficient and Scalable Deep Learning on Graphs. (2019).
  • Wang et al. (2020c) Wen Wang, Wei Zhang, Shukai Liu, Qi Liu, Bo Zhang, Leyu Lin, and Hongyuan Zha. 2020c. Beyond clicks: Modeling multi-relational item graph for session-based target behavior prediction. In Proceedings of The Web Conference 2020. 3056–3062.
  • Wang et al. (2021b) Wen Wang, Wei Zhang, Shukai Liu, Qi Liu, Bo Zhang, Leyu Lin, and Hongyuan Zha. 2021b. Incorporating Link Prediction into Multi-Relational Item Graph Modeling for Session-based Recommendation. IEEE Transactions on Knowledge and Data Engineering (2021).
  • Wang et al. (2019a) Xiang Wang, Xiangnan He, Yixin Cao, Meng Liu, and Tat-Seng Chua. 2019a. Kgat: Knowledge graph attention network for recommendation. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 950–958.
  • Wang et al. (2019b) Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019b. Neural graph collaborative filtering. In Proceedings of the 42nd international ACM SIGIR conference on Research and development in Information Retrieval. 165–174.
  • Wang et al. (2020a) Xiang Wang, Hongye Jin, An Zhang, Xiangnan He, Tong Xu, and Tat-Seng Chua. 2020a. Disentangled graph collaborative filtering. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 1001–1010.
  • Wang et al. (2019c) Xiang Wang, Dingxian Wang, Canran Xu, Xiangnan He, Yixin Cao, and Tat-Seng Chua. 2019c. Explainable reasoning over knowledge graphs for recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 5329–5336.
  • Wang and Wang (2014) Xinxi Wang and Ye Wang. 2014. Improving content-based and hybrid music recommendation using deep learning. In Proceedings of the 22nd ACM international conference on Multimedia. 627–636.
  • Wang et al. (2020b) Ziyang Wang, Wei Wei, Gao Cong, Xiao-Li Li, Xian-Ling Mao, and Minghui Qiu. 2020b. Global context enhanced graph neural networks for session-based recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 169–178.
  • Wilhelm et al. (2018) Mark Wilhelm, Ajith Ramanathan, Alexander Bonomo, Sagar Jain, Ed H Chi, and Jennifer Gillenwater. 2018. Practical diversified recommendations on youtube with determinantal point processes. In Proceedings of the 27th ACM International Conference on Information and Knowledge Management. 2165–2173.
  • Wu et al. (2020d) Fangzhao Wu, Ying Qiao, Jiun-Hung Chen, Chuhan Wu, Tao Qi, Jianxun Lian, Danyang Liu, Xing Xie, Jianfeng Gao, Winnie Wu, et al. 2020d. Mind: A large-scale dataset for news recommendation. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 3597–3606.
  • Wu et al. (2020a) Jiancan Wu, Xiangnan He, Xiang Wang, Qifan Wang, Weijian Chen, Jianxun Lian, and Xing Xie. 2020a. Graph Convolution Machine for Context-aware Recommender System. arXiv preprint arXiv:2001.11402 (2020).
  • Wu et al. (2021c) Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Chen, Jianxun Lian, and Xing Xie. 2021c. Self-supervised graph learning for recommendation. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 726–735.
  • Wu et al. (2021a) Le Wu, Lei Chen, Pengyang Shao, Richang Hong, Xiting Wang, and Meng Wang. 2021a. Learning Fair Representations for Recommendation: A Graph-based Perspective. In Proceedings of the Web Conference 2021. 2198–2208.
  • Wu et al. (2020b) Le Wu, Junwei Li, Peijie Sun, Richang Hong, Yong Ge, and Meng Wang. 2020b. Diffnet++: A neural influence and interest diffusion network for social recommendation. IEEE Transactions on Knowledge and Data Engineering (2020).
  • Wu et al. (2019a) Le Wu, Peijie Sun, Yanjie Fu, Richang Hong, Xiting Wang, and Meng Wang. 2019a. A neural influence diffusion model for social recommendation. In Proceedings of the 42nd international ACM SIGIR conference on research and development in information retrieval. 235–244.
  • Wu et al. (2018a) Le Wu, Peijie Sun, Richang Hong, Yong Ge, and Meng Wang. 2018a. Collaborative neural social recommendation. IEEE transactions on systems, man, and cybernetics: systems (2018).
  • Wu et al. (2018b) Peizhi Wu, Yi Tu, Xiaojie Yuan, Adam Jatowt, and Zhenglu Yang. 2018b. Neural Framework for Joint Evolution Modeling of User Feedback and Social Links in Dynamic Social Networks.. In IJCAI. 1632–1638.
  • Wu et al. (2019c) Qitian Wu, Hengrui Zhang, Xiaofeng Gao, Peng He, Paul Weng, Han Gao, and Guihai Chen. 2019c. Dual graph attention networks for deep latent representation of multifaceted social effects in recommender systems. In The World Wide Web Conference. 2091–2102.
  • Wu et al. (2021b) Shiwen Wu, Fei Sun, Wentao Zhang, and Bin Cui. 2021b. Graph neural networks in recommender systems: a survey. arXiv preprint arXiv:2011.02260 (2021).
  • Wu et al. (2019b) Shu Wu, Yuyuan Tang, Yanqiao Zhu, Liang Wang, Xing Xie, and Tieniu Tan. 2019b. Session-based recommendation with graph neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 346–353.
  • Wu et al. (2019d) Shu Wu, Mengqi Zhang, Xin Jiang, Xu Ke, and Liang Wang. 2019d. Personalizing graph neural networks with attention mechanism for session-based recommendation. arXiv preprint arXiv:1910.08887 (2019).
  • Wu et al. (2020c) Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. 2020c. A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning systems 32, 1 (2020), 4–24.
  • Xia et al. (2021a) Lianghao Xia, Chao Huang, Yong Xu, Peng Dai, Mengyin Lu, and Liefeng Bo. 2021a. Multi-Behavior Enhanced Recommendation with Cross-Interaction Collaborative Relation Modeling. In 2021 IEEE 37th International Conference on Data Engineering (ICDE). IEEE, 1931–1936.
  • Xia et al. (2021b) Lianghao Xia, Chao Huang, Yong Xu, Peng Dai, Xiyue Zhang, Hongsheng Yang, Jian Pei, and Liefeng Bo. 2021b.

    Knowledge-Enhanced Hierarchical Graph Transformer Network for Multi-Behavior Recommendation. In

    Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 4486–4493.
  • Xia et al. (2021c) Lianghao Xia, Yong Xu, Chao Huang, Peng Dai, and Liefeng Bo. 2021c. Graph Meta Network for Multi-Behavior Recommendation. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 757–766.
  • Xia et al. (2021d) Xin Xia, Hongzhi Yin, Junliang Yu, Yingxia Shao, and Lizhen Cui. 2021d. Self-Supervised Graph Co-Training for Session-based Recommendation. arXiv preprint arXiv:2108.10560 (2021).
  • Xia et al. (2021e) Xin Xia, Hongzhi Yin, Junliang Yu, Qinyong Wang, Lizhen Cui, and Xiangliang Zhang. 2021e. Self-Supervised Hypergraph Convolutional Networks for Session-based Recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 4503–4511.
  • Xian et al. (2019) Yikun Xian, Zuohui Fu, S. Muthukrishnan, Gerard de Melo, and Yongfeng Zhang. 2019. Reinforcement Knowledge Graph Reasoning for Explainable Recommendation. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR’19). Association for Computing Machinery, New York, NY, USA, 285–294.
  • Xiao et al. (2021) Fengtong Xiao, Lin Li, Weinan Xu, Jingyu Zhao, Xiaofeng Yang, Jun Lang, and Hao Wang. 2021. DMBGN: Deep Multi-Behavior Graph Networks for Voucher Redemption Rate Prediction. arXiv preprint arXiv:2106.03356 (2021).
  • Xie et al. (2016) Min Xie, Hongzhi Yin, Fanjiang Xu, Hao Wang, and Xiaofang Zhou. 2016. Graph-based metric embedding for next poi recommendation. In International Conference on Web Information Systems Engineering. Springer, 207–222.
  • Xie et al. (2021) Ruobing Xie, Qi Liu, Shukai Liu, Ziwei Zhang, Peng Cui, Bo Zhang, and Leyu Lin. 2021. Improving Accuracy and Diversity in Matching of Recommendation with Diversified Preference Network. arXiv preprint arXiv:2102.03787 (2021).
  • Xu et al. (2019c) Chengfeng Xu, Pengpeng Zhao, Yanchi Liu, Victor S Sheng, Jiajie Xu, Fuzhen Zhuang, Junhua Fang, and Xiaofang Zhou. 2019c. Graph Contextualized Self-Attention Network for Session-based Recommendation.. In IJCAI, Vol. 19. 3940–3946.
  • Xu et al. (2019a) Fengli Xu, Jianxun Lian, Zhenyu Han, Yong Li, Yujian Xu, and Xing Xie. 2019a. Relation-aware graph convolutional networks for agent-initiated social e-commerce recommendation. In Proceedings of the 28th ACM international conference on information and knowledge management. 529–538.
  • Xu et al. (2019b) Fengli Xu, Jianxun Lian, Zhenyu Han, Yong Li, Yujian Xu, and Xing Xie. 2019b. Relation-Aware Graph Convolutional Networks for Agent-Initiated Social E-Commerce Recommendation. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management (CIKM ’19). Association for Computing Machinery, New York, NY, USA, 529–538.
  • Xu et al. (2020) Huance Xu, Chao Huang, Yong Xu, Lianghao Xia, Hao Xing, and Dawei Yin. 2020. Global context enhanced social recommendation with hierarchical graph neural networks. In 2020 IEEE International Conference on Data Mining (ICDM). IEEE, 701–710.
  • Yang and Dong (2020) Zuoxi Yang and Shoubin Dong. 2020. HAGERec: hierarchical attention graph convolutional network incorporating knowledge graph for explainable recommendation. Knowledge-Based Systems 204 (2020), 106194.
  • Yao et al. (2019) Liang Yao, Chengsheng Mao, and Yuan Luo. 2019. Graph convolutional networks for text classification. In Proceedings of the AAAI conference on artificial intelligence, Vol. 33. 7370–7377.
  • Yao et al. (2018) Quanming Yao, Mengshuo Wang, Yuqiang Chen, Wenyuan Dai, Yu-Feng Li, Wei-Wei Tu, Qiang Yang, and Yang Yu. 2018. Taking human out of learning applications: A survey on automated machine learning. arXiv preprint arXiv:1810.13306 (2018).
  • Ying et al. (2018) Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L Hamilton, and Jure Leskovec. 2018. Graph convolutional neural networks for web-scale recommender systems. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 974–983.
  • Yu et al. (2021d) Bo Yu, Ruoqian Zhang, Wei Chen, and Junhua Fang. 2021d. Graph neural network based model for multi-behavior session-based recommendation. GeoInformatica (2021), 1–19.
  • Yu et al. (2020b) Feng Yu, Yanqiao Zhu, Qiang Liu, Shu Wu, Liang Wang, and Tieniu Tan. 2020b. TAGNN: Target attentive graph neural networks for session-based recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 1921–1924.
  • Yu et al. (2021a) Junliang Yu, Hongzhi Yin, Min Gao, Xin Xia, Xiangliang Zhang, and Nguyen Quoc Viet Hung. 2021a. Socially-Aware Self-Supervised Tri-Training for Recommendation. arXiv preprint arXiv:2106.03569 (2021).
  • Yu et al. (2020a) Junliang Yu, Hongzhi Yin, Jundong Li, Min Gao, Zi Huang, and Lizhen Cui. 2020a. Enhance social recommendation with adversarial graph convolutional networks. IEEE Transactions on Knowledge and Data Engineering (2020).
  • Yu et al. (2021b) Junliang Yu, Hongzhi Yin, Jundong Li, Qinyong Wang, Nguyen Quoc Viet Hung, and Xiangliang Zhang. 2021b. Self-Supervised Multi-Channel Hypergraph Convolutional Network for Social Recommendation. In Proceedings of the Web Conference 2021. 413–424.
  • Yu et al. (2021c) Junliang Yu, Hongzhi Yin, Jundong Li, Qinyong Wang, Nguyen Quoc Viet Hung, and Xiangliang Zhang. 2021c. Self-Supervised Multi-Channel Hypergraph Convolutional Network for Social Recommendation. In Proceedings of the Web Conference 2021. 413–424.
  • Zhang et al. (2019b) Chuxu Zhang, Dongjin Song, Chao Huang, Ananthram Swami, and Nitesh V Chawla. 2019b. Heterogeneous graph neural network. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 793–803.
  • Zhang et al. (2021a) Jun Zhang, Chen Gao, Depeng Jin, and Yong Li. 2021a. Group-Buying Recommendation for Social E-Commerce. In 2021 IEEE 37th International Conference on Data Engineering (ICDE). IEEE, 1536–1547.
  • Zhang et al. (2021b) Jiyong Zhang, Xin Liu, Xiaofei Zhou, and Xiaowen Chu. 2021b. Leveraging graph neural networks for point-of-interest recommendations. Neurocomputing 462 (2021), 1–13.
  • Zhang et al. (2019a) Jiani Zhang, Xingjian Shi, Shenglin Zhao, and Irwin King. 2019a. STAR-GCN: Stacked and Reconstructed Graph Convolutional Networks for Recommender Systems. In Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI’19). AAAI Press, 4264–4270.
  • Zhang et al. (2021c) Mengqi Zhang, Shu Wu, Xueli Yu, and Liang Wang. 2021c. Dynamic Graph Neural Networks for Sequential Recommendation. arXiv preprint arXiv:2104.07368 (2021).
  • Zhang et al. (2020) Weifeng Zhang, Jingwen Mao, Yi Cao, and Congfu Xu. 2020. Multiplex Graph Neural Networks for Multi-behavior Recommendation. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 2313–2316.
  • Zhang and Chen (2020) Yongfeng Zhang and Xu Chen. 2020. Explainable Recommendation: A Survey and New Perspectives. Foundations and Trends® in Information Retrieval 14, 1 (2020), 1–101. https://doi.org/10.1561/1500000066
  • Zhang et al. (2014) Yongfeng Zhang, Guokun Lai, Min Zhang, Yi Zhang, Yiqun Liu, and Shaoping Ma. 2014.

    Explicit factor models for explainable recommendation based on phrase-level sentiment analysis. In

    Proceedings of the 37th international ACM SIGIR conference on Research & development in information retrieval. 83–92.
  • Zhao et al. (2019a) Cheng Zhao, Chenliang Li, and Cong Fu. 2019a. Cross-domain recommendation via preference propagation graphnet. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 2165–2168.
  • Zhao et al. (2019b) Cheng Zhao, Chenliang Li, and Cong Fu. 2019b. Cross-domain recommendation via preference propagation graphnet. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 2165–2168.
  • Zheng et al. (2021a) Yu Zheng, Chen Gao, Liang Chen, Depeng Jin, and Yong Li. 2021a. DGCN: Diversified Recommendation with Graph Convolutional Networks. In Proceedings of the Web Conference 2021. 401–412.
  • Zheng et al. (2021b) Yu Zheng, Chen Gao, Xiangnan He, Depeng Jin, and Yong Li. 2021b. Incorporating Price into Recommendation with Graph Convolutional Networks. IEEE Transactions on Knowledge and Data Engineering (2021).
  • Zheng et al. (2020a) Yu Zheng, Chen Gao, Xiangnan He, Yong Li, and Depeng Jin. 2020a. Price-aware recommendation with graph convolutional networks. In 2020 IEEE 36th International Conference on Data Engineering (ICDE). IEEE, 133–144.
  • Zheng et al. (2020b) Yujia Zheng, Siyi Liu, Zekun Li, and Shu Wu. 2020b. DGTN: Dual-channel Graph Transition Network for Session-based Recommendation. In 2020 International Conference on Data Mining Workshops (ICDMW). IEEE, 236–242.
  • Zheng et al. (2021c) Zhi Zheng, Chao Wang, Tong Xu, Dazhong Shen, Penggang Qin, Baoxing Huai, Tongzhu Liu, and Enhong Chen. 2021c. Drug Package Recommendation via Interaction-aware Graph Induction. In Proceedings of the Web Conference 2021. 1284–1295.
  • Zhou et al. (2018) Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click-through rate prediction. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1059–1068.
  • Zhou et al. (2021) Huachi Zhou, Qiaoyu Tan, Xiao Huang, Kaixiong Zhou, and Xiaoling Wang. 2021. Temporal Augmented Graph Neural Networks for Session-Based Recommendations. (2021).
  • Zhou et al. (2020a) Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang, Changcheng Li, and Maosong Sun. 2020a. Graph neural networks: A review of methods and applications. AI Open 1 (2020), 57–81.
  • Zhou et al. (2020b) Kaixiong Zhou, Xiao Huang, Yuening Li, Daochen Zha, Rui Chen, and Xia Hu. 2020b. Towards Deeper Graph Neural Networks with Differentiable Group Normalization. Advances in Neural Information Processing Systems 33 (2020).
  • Zhu et al. (2019) Rong Zhu, Kun Zhao, Hongxia Yang, Wei Lin, Chang Zhou, Baole Ai, Yong Li, and Jingren Zhou. 2019. AliGraph: a comprehensive graph neural network platform. Proceedings of the VLDB Endowment 12, 12 (2019), 2094–2105.
  • Zhu et al. (2021) Tianyu Zhu, Leilei Sun, and Guoqing Chen. 2021. Graph-based Embedding Smoothing for Sequential Recommendation. IEEE Transactions on Knowledge and Data Engineering (2021).
  • Zhu et al. (2018) Ziwei Zhu, Xia Hu, and James Caverlee. 2018.

    Fairness-aware tensor-based recommendation. In

    Proceedings of the 27th ACM International Conference on Information and Knowledge Management. 1153–1162.
  • Zhuang et al. (2018) Tao Zhuang, Wenwu Ou, and Zhirong Wang. 2018. Globally optimized mutual influence aware ranking in e-commerce search. arXiv preprint arXiv:1805.08524 (2018).