The Role of Shopping Mission in Retail Customer Segmentation

09/06/2019 ∙ by Ondřej Sokol, et al. ∙ University of Economics, Prague (Vysoká škola ekonomická v Praze) 0

In retailing, it is important to understand customer behavior and determine customer value. A useful tool to achieve this goal is the cluster analysis of transaction data. Typically, a customer segmentation is based on the recency, frequency and monetary value of shopping or the structure of purchased products. We take a different approach and base our segmentation on a shopping mission - a reason why a customer visits the shop. Shopping missions include emergency purchases of specific product categories and general purchases of various sizes. In an application to a Czech drugstore chain, we show that the proposed segmentation brings unique information about customers and should be used alongside the traditional methods.

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

Retail chains have a huge amount of sales data available. An analysis of these data strives to understand the customer behavior and determine the customer value in order to increase profits. The information about customers can be utilized to increase the efficiency of the new product development (Li et al., 2012), product positioning (Gruca and Klemz, 2003), cross-category dependence (Hruschka et al., 1999; Russell and Petersen, 2000; Leeflang et al., 2008), product complements and substitutes determination (Srivastava et al., 1981; Chib et al., 2002), promotions planning (Trappey et al., 2009), online marketing (Chen et al., 2009), targeted advertising (Jonker et al., 2004; Zhang et al., 2007), product recommendation (Liu and Shih, 2005), product association rules (Weng, 2016), stock optimization (Borin et al., 1994) and other retail operations. One of the tools used to achieve these goals is the cluster analysis.

There are many applications of the cluster analysis in retail business. Products sold by the shop can be clustered according to their characteristics in order to find substitutes and complements (Srivastava et al., 1981) or target market (Zhang et al., 2007). A product categorization based solely on customer shopping patterns was proposed by Holý et al. (2017). Customers can be segmented according to their demographics and lifestyle or their shopping behavior. A popular approach is to segment customers based on the recency, frequency and monetary value (RFM) of their shopping (Kahan, 1998; Miglautsch, 2000; Yang, 2004; Chen et al., 2009; Khajvand and Tarokh, 2011; Putra et al., 2012; Peker et al., 2017). Another approach is to segment customers based on the purchased products structure (PPS) (Russell and Kamakura, 1997; Manchanda et al., 1999; Andrews and Currim, 2002; Tsai and Chiu, 2004). Lingras et al. (2014) and Ammar et al. (2016) simultaneously clustered both products and customers. Besides traditional retail business, customer segmentation can also be utilized in e-commerce (Ballestar et al., 2018; Dahana et al., 2019) and the sharing economy (Lutz and Newlands, 2018). Overall, the cluster analysis brings useful insight into the customer behavior and helps in the decision-making process, especially when combined with the business knowledge (Seret et al., 2014).

We deal with the customer segmentation using data from receipts. The traditional RFM and PPS segmentations answer the questions:

  • When was the last time customers visited the shop?

  • How often do customers visit the shop?

  • How much money do customers spend?

  • What product categories do customers buy?

In our analysis, we propose to segment customers based on their shopping mission (SM). The proposed segmentation answers the questions:

  • What is the purpose of customer visits?

  • Do customers visit the shop because of a specific product category?

  • Do customers buy products in other shops?

The proposed approach brings a new insight into the structure of customers. As a result, it can be used in many marketing areas such as promotion planing, shelf management, improvement of customer loyalty as well as general prediction of sales.

The main idea of the proposed approach is as follows. We utilize the transaction data in a form of receipts which are linked to customers through the loyalty program. We first cluster individual baskets and then use this information to segment customers. This is illustrated in Figure 1 along with a comparison to the PPS segmentation. The k-means method is utilized for both the basket clustering and the customer segmentation. An analysis of a Czech drugstore chain shows that there are some customers who visit the shop due to an emergency purchase of a specific product category while others prefer general purchase. Another segmentation based on a shopping mission was presented by Reutterer et al. (2006). The main difference from our proposed method is that we also consider the value of the basket. The proposed method is not meant to replace the RFM or PPS segmentation but rather to be used alongside them and to bring a new perspective. The combination of RFM, PPS and SM approaches forms a versatile segmentation based on a broad range of customer characteristics not tied to a single specific purpose. We emphasize the interpretability and usability by marketing departments and other experts involved in the retail decision-making process.

The rest of the paper is structured as follows. In Section 2, we describe the general structure of transaction data in retail business. In Section 3, we review the segmentation based on the recency, frequency, and value of a shopping with an application to our dataset. In Section 3, we review the segmentation based on the structure of purchased products and again apply it to our dataset. In Section 5, we propose a novel segmentation based on the shopping mission of a customer with an application to our dataset. In Section 6, we show how all three segmentations can be combined. We conclude the paper in Section 7.

Figure 1: The process of the PPS segmentation and the SM segmentation of customers.

2 Transaction Data

We perform the analysis of retail business using the transaction data. The hierarchical structure of these data is illustrated in Figure 1.

A product is characterized by the brand, physical properties, and purpose. Note that the price of the product and whether the product is in sales promotion can vary over time and therefore we put it to the receipt data. Because there are many products, it is useful to aggregate them into product categories. We denote the product categories as . An individual product bought by the customer is referred to as the purchased product. We denote the purchased products as . Each purchased product belongs to a single product category.

A purchased basket is a set of purchased products. We denote the purchased baskets as . A specific purchased basket is a subset of all purchased products, i.e. , . A receipt is a purchased basket with additional information about the customer ID, prices, sales promotion, date and time.

A customer history is a set of purchased baskets. We denote the customer history as . A specific customer history is a subset of all purchased baskets, i.e. , . A customer is a customer history with additional information about the contact, gender, age, number of children and other demographic information.

In the paper, we analyze a sample of real data. Our dataset consists of individual purchase data of one of the retail chains in the drugstore market of the Czech Republic. We use a three-month dataset of receipts which include more than 5.6 million baskets bought by more than 1.5 million customers with the loyalty card. Each row in a receipt stands for a single purchased product. The retail chain sells over 10 thousand products which are divided into 55 categories based on their purpose. This categorization was done by an expert opinion.

3 Segmentation Based on Recency, Frequency and Monetary Value

One of the most popular way to segment customers is the clustering using data about recency, frequency and monetary value (RFM) of their shopping. This method is fast and simple as its original purpose is to provide an easy-to-implement framework for quantifying customer behavior (Kahan, 1998; Miglautsch, 2000).

Yang (2004) described some shortages of RFM method (e.g. the inability of the RFM to generate the real differences among RFM cells) and introduced a single predictor which is consolidated from the three variables of RFM. A method for the sequential pattern mining using RFM segmentation was presented by Chen et al. (2009). Khajvand and Tarokh (2011)

improve RFM segmentation by using the adapted RFM in order to estimate the customer lifetime value. Expansion of RFM segmentation by the fusion with ART2 algorithm to cluster the customers in the retail company was presented by

Putra et al. (2012). Another expansion of RFM by Peker et al. (2017) proposes to include customer relation length and periodicity to the customer segmentation.

The first RFM characteristic of a customer is the recency (Rec). Customers are segmented by the time of the last purchase occurrence. With the knowledge of the recency of the last purchase, the retailer can use different marketing techniques to attract customers who had been in the shop in the last week and customers who had not been there for months. In the broader concept, analysis of the occurrences of shopping in time can help for example to identify leaving customers, i.e. those who used to visit the shop frequently but their shopping behavior changed in the recent history. The goal of the marketing department is to prevent customers from leaving completely. Similarly, a customer who is in the shop for the first time ever may require special attention in order to keep him. Under the assumption that a finite number of customers traits exists based on the recency which do not change in time, a customer , is assigned to a recency segment , according to the variable

(1)

The centers of segments are either found by some clustering algorithm or defined by the expert, which is more common. The typical number of segments is around 5. A small number of segments gives an easy and useful interpretation of customers assigned to each given segment.

The second RFM characteristic of a customer is the frequency (Frq). The purchase frequency is defined as a number of visits of a customer during given time frame. Along with the average value of a basket, it is one of the most tracked key performance indicators. The marketing department can use the information about frequency and distinguish loyal customers from the ones who go to the shop just in a case of emergency. While the goal for the loyal customers is to preserve their shopping behavior, the customers who rarely visit the shop should be recruited. A customer , is assigned to a frequency segment , according to the variable

(2)

As in the case of the recency, the centers of segments are either found by some clustering algorithm or defined by the expert. The typical number of segments is around 5.

The third RFM characteristic of a customer is the monetary value (Mon). A customer segmentation by monetary value can be done in various ways. A common approach is to compute either the sums of all sales during a given time frame or the average value of baskets in a given time frame for each customer. The latter is used in a combination with the frequency analysis. Retailers can also focus on margins instead of sales. In our case, we assign a customer , to a monetary value segment , according to the variable

(3)

Again, the centers of segments are either found by some clustering algorithm or defined by the expert. The typical number of segments is from 5 to 10.

The combination of the above mentioned approaches forms the RFM segmentation. One approach to derive RFM segmentation is to create the 3-dimensional matrix of all combinations of the , and segmentations with size . However, the number of clusters can be quite large. To reduce the number of segments, another approach may be used. For a given , a customer , can be assigned to a RFM segment ,

according to the vector variable

(4)

The centers of these segments are found by some clustering algorithms such as the -means method. Despite its shortages, the RFM segmentation is commonly used across retail business for its simplicity and straightforward interpretation.

4 Segmentation Based on Purchased Products Structure

Products in retail shops are often categorized based on their properties such as a purpose, price, pack size and brand. A basic approach is to use product purpose to deliver product category. The categorization of products can be done either by an expert or by an algorithm (Srivastava et al., 1981; Zhang et al., 2007; Holý et al., 2017). Subsequently, customers can be segmented using their receipts. We refer to this clustering as purchased product structure (PPS) segmentation. The knowledge of their purchases is directly linked to product categories. Such analysis reveals commonly bought categories and therefore helps in targeting of marketing campaigns.

Segmentation of customers based on their category purchases was studied in Russell and Kamakura (1997), where authors segmented customers with respect to brand preference using household purchase data. Another approach of using product categorization on household data to analyze customers behavior was published by Manchanda et al. (1999)

. A method for identifying customer segments with identical choice behaviors across product categories using logit model was presented by

Andrews and Currim (2002). Tsai and Chiu (2004) dealt with clustering customers based on their purchase data linked to product categories and presented a methodology to ensure the quality of the resulting clustering. Lingras et al. (2014) and Ammar et al. (2016) utilized an iterative meta-clustering technique that uses clustering results from one set of objects to dynamically change the representation of another set of objects. The method is applied on product categorization and customer segmentation using supermarket basket data.

We segment customers based on ratios of their purchases in each category. For a customer , , the PPS segmentation is based on information about product category , given by

(5)

A customer , is then assigned to a PPS segment , according to the vector variable

(6)

The centers of PPS segments are found using the k-means method. The optimal number of clusters

is chosen according to the ratio of between cluster variance and total variance and Davies-Bouldin index alongside with a reasonable interpretation of resulting clusters.

We perform a customer segmentation in a Czech drugstore chain according to 55 product categories. We find that in our case the optimal number of clusters is 12. Figure 2 shows that there are 11 specialized segments and 1 general segment. The centers of specialized segments are composed of about 60% spendings in a single category. On the other hand, the general segment is fairly uniformly composed of over 15 popular categories. The distribution of customers assigned to the individual segments is alongside with labels of the dominant product categories shown in Table 1. It is worth noting that despite having 55 categories, top 15 categories comprise of over 98% of the total revenue.

Cluster Description Share of customers
P01 General 32.0%
P02 Specialized – Detergents 6.1%
P03 Specialized – Laundry detergents 5.6%
P04 Specialized – Body products 6.1%
P05 Specialized – Face products 6.2%
P06 Specialized – Dental products 5.7%
P07 Specialized – Hair products 7.5%
P08 Specialized – Beauty products 9.6%
P09 Specialized – Products for men 5.3%
P10 Specialized – Products for children 5.0%
P11 Specialized – Parfumes 7.3%
P12 Specialized – Seasonal products 3.7%
Table 1: Distribution of the PPS customer segmentation.
Figure 2: Ratios of product value from the PPS customer segments split into the product categories.

We use the PPS segmentation as a base customer clustering according to categories they purchase. The next step is to compare it with a more complex approach featuring an intermediate step and adding the basket information to the segmentation. Figure 1 shows the process of both segmentations.

5 Segmentation Based on Shopping Mission

We propose an addition to the RFM and PPS segmentations. The above mentioned segmentations lack the information about the reason why customers visit the shop. Some customers visit the shop just to buy one product they need. This is called the emergency purchase. Other customers purchase more different products. This is called the general purchase. Our goal is to estimate what is the reason why the customers come to the shop, i.e. what is their shopping mission

. From the marketing point of view, customers who come to the shop just for emergency reasons probably buy everything else in some other shop, therefore the goal is to transform them into regular customers. On the other hand, customers who already fulfill a majority of their needs in the shop are the most valuable and the goal of the marketing department is to retain them.

In the literature, the shopping mission or shopping motivation is often approached from a qualitative point of view. Hedonic shopping motivation and its effect in utilitarian enviroments was studied by Yim et al. (2014) using a field survey. Studies based on transaction data are present in the literature as well. Schröder (2017) analyzed multi-category purchase decisions on the weekly basis using the item response theory models which allows to reveal characteristics of households for purchase decisions. Underlying latent activities of shoppers are also focus of the study by Hruschka (2014)

using topic models. Analysis of baskets using self-organizing maps was presented by

Decker and Monien (2003). Reutterer et al. (2006) introduced a two-stage method of clustering customers using the basket clustering. In the first phase, baskets are clustered based on the purchased products, this is done using information whether the product appeared in the basket or not. In the second phase the customers are segmented based on their baskets. A method for identifying shopping mission using basket value and variety was proposed in Sarantopoulos et al. (2016).

To segment customers, we use the ratio of product categories in the basket along with the value of the basket. The main difference from the work of Reutterer et al. (2006) is the use of basket value, which is important in differentiating the emergency basket from the general one. It is important to note that we intend to use this segmentation alongside the others and do not try to replace any of the above mentioned segmentation. Therefore, we do not need to incorporate information about the frequency of shopping or total value as it is already involved in the RFM approach. We neither use the information about the total expenditure within the purpose categories as it is involved in the PPS approach. Our goal is simply to estimate the shopping mission of customers which may differ greatly from the RFM and PPS approaches.

The main reasoning behind including purpose categories as well as the basket value is to get easily interpretable clusters of baskets. In order to get reasonable clustering, the value of basket is normalized using the 95% quantile of a basket value while the baskets with the value over this quantile are set to 1 due to a skewed distribution of the basket value. See Figure

3 for the kernel density function of the normalized basket value. In the clustering, each basket is then represented by a vector of non-negative ratios with unit sum and a normalized basket value ranging from 0 to 1. The interpretation is that we give similar weights to both the structure and the value of the basket. For a basket , , the PPS clustering is based on information about product category , given by

(7)

and information about value given by

(8)

where is the 95% quantile of all basket values. A basket , is then assigned to a SM cluster , according to the vector variable

(9)

The centers of SM basket segments are found using the k-means method and the optimal number of clusters is chosen according to the ratio of between cluster variance and total variance and Davies-Bouldin index.

Figure 3: Estimated kernel density function of normalized basket value.

Our basket segmentation have the following geometric interpretation. Let us denote

(10)

For a given basket , then represents a point in a simplex of dimension while adds a depth to this simplex. The clustering is simply a dissectioning of this space. For simplicity, we focus on a low dimension of three product categories. The ratio of spending in a product category is then represented by a point in a triangle, whose vertices represent the exclusivity of a category in the basket. The value of basket adds a depth to the triangle and forms a prism. Each basket is a point in this space. The centers of resulting clusters are inside the prism as well. In Figure 4, we show two cuts of the prism with 4 cluster centers and defined cluster area for low and high basket value. The center C1 represents a point with a higher value than the others. Therefore its cluster area in cuts by basket value expands with a higher value of the basket. The baskets are assigned to the nearest center using the standard Euclidean distance.

Figure 4: Illustration of the SM segmentation for different basket value levels.

We cluster baskets in a Czech drugstore chain according to 55 product categories. As expected the basket clusters are formed around previously mentioned well-selling categories. We find that 12 is the optimal number of clusters. Two clusters represent small and big universal baskets with no dominant category while the 10 others are focused on a single dominant category. The between cluster variance ratio is 0.8 with 12 clusters while Davies-Bouldin index for 12 clusters has similar value to the other possible choices. The resulting distribution of baskets to the cluster as well as the interpretation of each cluster is shown in Table 2. Each cluster is named after the dominant category similarly to the PPS segmentation. The structure of each basket cluster (archetypes) is shown in Figure 5. It is evident that general baskets tend to have a higher value than the others.

Cluster Description Share of baskets
B01 General – Big 13.0%
B02 General – Small 13.9%
B03 Specialized – Detergents 9.8%
B04 Specialized – Laundry detergents 7.9%
B05 Specialized – Body products 6.5%
B06 Specialized – Face products 6.0%
B07 Specialized – Dental products 8.8%
B08 Specialized – Hair products 11.3%
B09 Specialized – Beauty products 9.7%
B10 Specialized – Products for men 2.2%
B11 Specialized – Products for children 6.4%
B12 Specialized – Feminine hygiene products 4.5%
Table 2: Distribution of the SM basket clustering.
Figure 5: Ratios of product value from the SM basket clusters split into the product categories.

In the second step, we determine the customer segments based on the ratio of baskets archetypes they bought. We do not use the absolute number of the baskets because of purely practical reasons. Our goal is to estimate the shopping mission of ordinary customers. However, some people visit the retail chain to supply their own small business. They buy an enormous number of products with a huge frequency. Clustering algorithms, in that case, are likely to create numerous clusters just for a very small number of customers. This is a logical and right way. However, this information about the value and frequency is already included in the RFM segmentation. Therefore we normalize the number of baskets by using the ratio of baskets archetypes bought by a customer. Customers with unusual shopping behavior are easily detectable using the RFM and SM segmentations as a whole, so we do not need to exclude them at all. For a customer , , the SM segmentation is based on information about basket cluster , given by

(11)

A customer , is then assigned to a SM segment , according to the vector variable

(12)

For the second phase we also use the k-means algorithm and select the optimal number of clusters according to the Davies-Bouldin index and ratio of between cluster variance.

We continue with our empirical analysis and segment customers of a Czech drugstore chain. We find the optimal number of clusters to be 18 using the between cluster variance ratio statistics. For the description of each segment, we use its center ratios of each basket type in the customer history. This allows us to distinguish three main types of customers. The general customers buy variety of categories in their baskets. As a customer with bulk purchases has a significantly different shopping motivation than a customer with very small yet various purchases, the general group is further divided into more segments based on the prevailing purchase size. The emergency customers focus only on one type of category in each of their purchases and visit the store with a very straightforward motivation. They are looking for specific products and are not willing to extend their purchase. The proposed segmentation further divide emergency customers into segments based on the category they prefer in majority of their purchases. The mixed customers are a combination of the above customer types. Overall, the clustering consists of 5 general segments with different basket values, 12 emergency segments formed around a single basket type and 3 mixed segments of both general and emergency baskets. The interpretation of the segments along with the percentage of assigned customers is shown in Table 3. The structure of the baskets archetypes in clusters is shown in Figure 6.

Cluster Description Share of customers
M01 General – Exclusively small 6.0%
M02 General – Mainly small 8.8%
M03 General – Small and big 8.4%
M04 General – Mainly big 11.3%
M05 General – Exclusively big 6.8%
M06 Emergency – Detergents 2.9%
M07 Emergency – Laundry detergents 4.3%
M08 Emergency – Body products 3.3%
M09 Emergency – Face products 3.6%
M10 Emergency – Dental products 3.1%
M11 Emergency – Hair products 4.2%
M12 Emergency – Beauty products 4.2%
M13 Emergency – Products for men 3.0%
M14 Emergency – Products for children 4.3%
M15 Emergency – Feminine hygiene products 1.4%
M16 Mixed – Detergents 8.2%
M17 Mixed – Hair products 9.6%
M18 Mixed – Beauty products 6.4%
Table 3: Distribution of the SM customer segmentation.
Figure 6: Ratios of baskets from the SM customer segments split into the SM basket clusters.

The division into the three main customer types and the subdivision into the specific segments is important in choosing suitable marketing strategies and is not contained in the common RFV and PPS segmentation techniques. Using the knowledge of experts in the field and the analysis of customer characteristics, each segment can be further described. For example, the segment of customers focused exclusively on the big baskets is distinguished by high proportion of promotion sales. Not surprisingly, the segment of customers focused on products for children are usually parents in their 20s and 30s. Such type of information is crucial for practical applications including marketing targeting and optimization of promotion sales. The proposed approach therefore offers a novel insight into the shopping behavior of customers.

6 Comparison of Segmentations

First, we compare the RFM, PPS and SM segmentations. Our goal is to find if the segmentations are similar or if each segmentation brings unique information to the customer analysis. We adopt the purity measure for comparison. Let us assume we have objects clustered by methods and with and clusters. The purity is then defined as

(13)

where is the set of objects in the cluster of the method and is the set of objects in the cluster of the method . Similar clusterings have the purity close to 1 while different clusterings have the purity close to 0. Note that the purity is not symmetrical. Table 4 reports the purity for segmentations based on recency (Rec), frequency (Frq), monetary value (Mon), purchased product structure (PPS) and shopping mission (SM). We can see that each segmentation is unique as there are no segmentations with a high similarity. However, a medium similarity does exist. For example, the F and PPS segmentations are related to the SM approach due to Frq/SM purity and PPS/SM purity . Reverse relationships have much lower purities because the SM segmentation has significantly more clusters than other segmentations.

Rec Frq Mon PPS SM
Rec 1.000 0.298 0.208 0.189 0.270
Frq 0.484 1.000 0.403 0.323 0.523
Mon 0.325 0.372 1.000 0.272 0.311
PPS 0.433 0.416 0.416 1.000 0.428
SM 0.183 0.184 0.131 0.129 1.000
Table 4: Purity between customer segmentations in rows and columns.

Next, we investigate the relationship between the SM and Frq segmentations in more detail. We compare 6 Frq segments and SM segments described in Table 3. Figure 7 shows how customer segments from the SM approach are divided into the Frq segments. The general segments M02–M04 and mixed segments M16–M18 have relatively high frequency while general segments M01 and M05 and the specialized emergency segments M05–M15 have quite low frequency. This is an expected result as loyal customers with a general shopping visit the shop more often than customers that mainly shop elsewhere and visit the shop only for emergencies.

Figure 7: Ratios of customers from the SM segments split into the Frq segments.

Finally, we investigate the relationship between the SM and PPS segmentations in more detail. We compare PPS segments described in Table 1 and SM segments described in Table 3. Figure 8 shows how customer segments from the SM approach are divided into the PPS segments. We can see that specialized segments P02–P10 correspond to segments M05–M14. General segment P01 is split among M01–M05 clusters according to the value of typical baskets and to clusters M16–M18 with dominant products. Interestingly, feminine hygiene products form their own segment M15. This is because many customers visit the drugstore specifically for these products but also purchase them in general baskets. Specialized segments P11 and P12 are clustered into general segments M01–M05. This is because customers do not visit shop specifically for these products but rather buy them together with other products.

Figure 8: Ratios of customers from the SM segments split into the PPS segments.

All considered segmentations should be used together in the analysis of customers as each segmentation has a unique structure and interpretation and brings different information about customers.

7 Conclusion

We deal with a segmentation of customers in retail business according to their shopping behavior. The paper has two main contributions.

  • First, we propose a new segmentation approach based on a shopping mission of a customer. The shopping mission answers the question why the customer visits the shop. Possible shopping missions include the emergency purchase of a specific product category and the general purchase.

  • Second, we show how various segmentations can be combined in a real application. Besides the proposed method, we also consider recency, frequency and monetary value approach as well as the approach based on the structure of purchased products. The results show that the proposed segmentation brings useful insight into the analysis of customer behavior.

The proposed segmentation was introduced in a major Czech drugstore chain and is currently used mainly in e-mail targeting. The customer reaction indicators in targeted emailing campaigns such as the open rate and click rate have significantly improved in comparison to the previously used RFV and PPS segmentations.

Acknowledgements

We would like to thank Michal Černý for his comments and Alena Holá and Zuzana Veselá for proofreading.

Funding

This work was supported by the Internal Grant Agency of the University of Economics, Prague under Grant F4/21/2018.

References

  • Ammar et al. (2016) Ammar, A., Elouedi, Z., Lingras, P. 2016. Meta-Clustering of Possibilistically Segmented Retail Datasets. Fuzzy Sets and Systems. Volume 286. Pages 173–196. ISSN 0165-0114. {https://doi.org/10.1016/j.fss.2015.07.019}.
  • Andrews and Currim (2002) Andrews, R. L., Currim, I. S. 2002. Identifying Segments with Identical Choice Behaviors Across Product Categories: An Intercategory Logit Mixture Model. International Journal of Research in Marketing. Volume 19. Issue 1. Pages 65–79. ISSN 0167-8116. {https://doi.org/10.1016/S0167-8116(02)00048-4}.
  • Ballestar et al. (2018) Ballestar, M. T., Grau-Carles, P., Sainz, J. 2018. Customer Segmentation in E-Commerce: Applications to the Cashback Business Model. Journal of Business Research. Volume 88. Pages 407–414. ISSN 0148-2963. {https://doi.org/10.1016/j.jbusres.2017.11.047}.
  • Borin et al. (1994) Borin, N., Farris, P. W., Freeland, J. R. 1994. A Model for Determining Retail Product Category Assortment and Shelf Space Allocation. Decision Sciences. Volume 25. Issue 3. Pages 359–384. ISSN 0011-7315. {https://doi.org/10.1111/j.1540-5915.1994.tb01848.x}.
  • Chen et al. (2009) Chen, Y.-L., Kuo, M.-H., Wu, S.-Y., Tang, K. 2009. Discovering Recency, Frequency, and Monetary (RFM) Sequential Patterns from Customers’ Purchasing Data. Electronic Commerce Research and Applications. Volume 8. Issue 5. Pages 241–251. ISSN 15674223. {https://doi.org/10.1016/j.elerap.2009.03.002}.
  • Chib et al. (2002) Chib, S., Seetharaman, P. B., Strijnev, A. 2002. Analysis of Multi-Category Purchase Incidence Decisions Using IRI Market Basket Data. Advances in Econometrics. Volume 16. Pages 57–92. ISSN 0731-9053. {https://doi.org/10.1016/S0731-9053(02)16004-X}.
  • Dahana et al. (2019) Dahana, W. D., Miwa, Y., Morisada, M. 2019. Linking Lifestyle to Customer Lifetime Value: An Exploratory Study in an Online Fashion Retail Market. Journal of Business Research. Volume 99. Pages 319–331. ISSN 0148-2963. {https://doi.org/10.1016/j.jbusres.2019.02.049}.
  • Decker and Monien (2003) Decker, R., Monien, K. 2003. Market Basket Analysis with Neural Gas Networks and Self-Organising maps. Journal of Targeting, Measurement and Analysis for Marketing. Volume 11. Issue 4. Pages 373–386. ISSN 0967-3237. {https://doi.org/10.1057/palgrave.jt.5740092}.
  • Gruca and Klemz (2003) Gruca, T. S., Klemz, B. R. 2003.

    Optimal New Product Positioning: A Genetic Algorithm Approach.

    European Journal of Operational Research. Volume 146. Issue 3. Pages 621–633. ISSN 0377-2217. {https://doi.org/10.1016/s0377-2217(02)00349-1}.
  • Holý et al. (2017) Holý, V., Sokol, O., Černý, M. 2017. Clustering Retail Products Based on Customer Behaviour. Applied Soft Computing. Volume 60. Pages 752–762. ISSN 1568-4946. {https://doi.org/10.1016/j.asoc.2017.02.004}.
  • Hruschka (2014) Hruschka, H. 2014. Linking Multi-Category Purchases to Latent Activities of Shoppers: Analysing Market Baskets by Topic Models. Marketing: ZFP - Journal of Research and Management. Volume 36. Issue 4. Pages 267–278. ISSN 0344-1369. {https://doi.org/10.15358/0344-1369_2014_4_267}.
  • Hruschka et al. (1999) Hruschka, H., Lukanowicz, M., Buchta, C. 1999. Cross-Category Sales Promotion Effects. Journal of Retailing and Consumer Services. Volume 6. Issue 2. Pages 99–105. ISSN 0969-6989. {https://doi.org/10.1016/S0969-6989(98)00026-5}.
  • Jonker et al. (2004) Jonker, J.-J., Piersma, N., Van den Poel, D. 2004. Joint Optimization of Customer Segmentation and Marketing Policy to Maximize Long-Term Profitability. Expert Systems with Applications. Volume 27. Issue 2. Pages 159–168. ISSN 0957-4174. {https://doi.org/10.1016/j.eswa.2004.01.010}.
  • Kahan (1998) Kahan, R. 1998. Using Database Marketing Techniques to Enhance Your One-to-One Marketing Initiatives. Journal of Consumer Marketing. Volume 15. Issue 5. Pages 491–493. ISSN 0736-3761. {https://doi.org/10.1108/07363769810235965}.
  • Khajvand and Tarokh (2011) Khajvand, M., Tarokh, M. J. 2011. Estimating Customer Future Value of Different Customer Segments Based on Adapted RFM Model in Retail Banking Context. Procedia Computer Science. Volume 3. Pages 1327–1332. ISSN 1877-0509. {https://doi.org/10.1016/j.procs.2011.01.011}.
  • Leeflang et al. (2008) Leeflang, P. S. H., Parreño Selva, J., Van Dijk, A., Wittink, D. R. 2008. Decomposing the Sales Promotion Bump Accounting for Cross-Category Effects. International Journal of Research in Marketing. Volume 25. Issue 3. Pages 201–214. ISSN 0167-8116. {https://doi.org/10.1016/j.ijresmar.2008.03.003}.
  • Li et al. (2012) Li, Y. L., Tang, J. F., Chin, K. S., Luo, X. G., Han, Y. 2012. Rough Set-Based Approach for Modeling Relationship Measures in Product Planning. Information Sciences. Volume 193. Pages 199–217. ISSN 0020-0255. {https://doi.org/10.1016/j.ins.2011.12.017}.
  • Lingras et al. (2014) Lingras, P., Elagamy, A., Ammar, A., Elouedi, Z. 2014. Iterative Meta-Clustering Through Granular Hierarchy of Supermarket Customers and Products. Information Sciences. Volume 257. Pages 14–31. ISSN 0020-0255. {https://doi.org/10.1016/j.ins.2013.09.018}.
  • Liu and Shih (2005) Liu, D. R., Shih, Y. Y. 2005. Integrating AHP and Data Mining for Product Recommendation Based on Customer Lifetime Value. Information and Management. Volume 42. Issue 3. Pages 387–400. ISSN 0378-7206. {https://doi.org/10.1016/j.im.2004.01.008}.
  • Lutz and Newlands (2018) Lutz, C., Newlands, G. 2018. Consumer Segmentation Within the Sharing Economy: The Case of Airbnb. Journal of Business Research. Volume 88. Pages 187–196. ISSN 0148-2963. {https://doi.org/10.1016/j.jbusres.2018.03.019}.
  • Manchanda et al. (1999) Manchanda, P., Ansari, A., Gupta, S. 1999. The "Shopping Basket": A Model for Multicategory Purchase Incidence Decisions. Marketing Science. Volume 18. Issue 2. Pages 95–114. ISSN 0732-2399. {https://doi.org/10.2307/19321195}.
  • Miglautsch (2000) Miglautsch, J. R. 2000. Thoughts on RFM Scoring. Journal of Database Marketing & Customer Strategy Management. Volume 8. Issue 1. Pages 67–72. ISSN 1741-2439. {https://doi.org/10.1057/palgrave.jdm.3240019}.
  • Peker et al. (2017) Peker, S., Kocyigit, A., Eren, P. E. 2017. LRFMP Model for Customer Segmentation in the Grocery Retail Industry: A Case Study. Marketing Intelligence & Planning. Volume 35. Issue 4. Pages 544–559. ISSN 0263-4503. {https://doi.org/10.1108/mip-11-2016-0210}.
  • Putra et al. (2012) Putra, I. K. G. D., Cahyawan, A. A. K. A., Shavitri H., D. 2012. Combination of Adaptive Resonance Theory 2 and RFM Model for Customer Segmentation in Retail Company. International Journal of Computer Applications. Volume 48. Issue 2. Pages 18–23. ISSN 0975-8887. {https://doi.org/10.5120/7320-0110}.
  • Reutterer et al. (2006) Reutterer, T., Mild, A., Natter, M., Taudes, A. 2006. A Dynamic Segmentation Approach for Targeting and Customizing Direct Marketing Campaigns. Journal of Interactive Marketing. Volume 20. Issue 3-4. Pages 43–57. ISSN 1094-9968. {https://doi.org/10.1002/dir.20066}.
  • Russell and Kamakura (1997) Russell, G. J., Kamakura, W. A. 1997. Modeling Multiple Category Brand Preference with Household Basket Data. Journal of Retailing. Volume 73. Issue 4. Pages 439–461. ISSN 0022-4359. {https://doi.org/10.1016/s0022-4359(97)90029-4}.
  • Russell and Petersen (2000) Russell, G. J., Petersen, A. 2000. Analysis of Cross Category Dependence in Market Basket Selection. Journal of Retailing. Volume 76. Issue 3. Pages 367–392. ISSN 0022-4359. {https://doi.org/10.1016/s0022-4359(00)00030-0}.
  • Sarantopoulos et al. (2016) Sarantopoulos, P., Theotokis, A., Pramatari, K., Doukidis, G. 2016. Shopping Missions: An Analytical Method for the Identification of Shopper Need States. Journal of Business Research. Volume 69. Issue 3. Pages 1043–1052. ISSN 0148-2963. {https://doi.org/10.1016/j.jbusres.2015.08.017}.
  • Schröder (2017) Schröder, N. 2017. Using Multidimensional Item Response Theory Models to Explain Multi-Category Purchases. Marketing: ZFP – Journal of Research and Management. Volume 39. Issue 2. Pages 27–37. ISSN 0344-1369. {https://doi.org/10.15358/0344-1369-2017-2-27}.
  • Seret et al. (2014) Seret, A., Verbraken, T., Baesens, B. 2014. A New Knowledge-Based Constrained Clustering Approach: Theory and Application in Direct Marketing. Applied Soft Computing. Volume 24. Pages 316–327. ISSN 1568-4946. {https://doi.org/10.1016/j.asoc.2014.06.002}.
  • Srivastava et al. (1981) Srivastava, R. K., Leone, R. P., Shocker, A. D. 1981.

    Market Structure Analysis: Hierarchical Clustering of Products Based on Substitution-in-Use.

    Journal of Marketing. Volume 45. Issue 3. Pages 38–48. ISSN 0022-2429. {https://doi.org/10.2307/1251540}.
  • Trappey et al. (2009) Trappey, C. V., Trappey, A. J. C., Chang, A.-C., Huang, A. Y. L. 2009. The Analysis of Customer Service Choices and Promotion Preferences Using Hierarchical Clustering. Journal of the Chinese Institute of Industrial Engineers. Volume 26. Issue 5. Pages 367–376. ISSN 1017-0669. {https://doi.org/10.1080/10170660909509151}.
  • Tsai and Chiu (2004) Tsai, C. Y., Chiu, C. C. 2004. A Purchase-Based Market Segmentation Methodology. Expert Systems with Applications. Volume 27. Issue 2. Pages 265–276. ISSN 0957-4174. {https://doi.org/10.1016/j.eswa.2004.02.005}.
  • Weng (2016) Weng, C.-H. 2016. Identifying Association Rules of Specific Later-Marketed Products. Applied Soft Computing. Volume 38. Pages 518–529. ISSN 1568-4946. {https://doi.org/10.1016/j.asoc.2015.09.047}.
  • Yang (2004) Yang, A. X. 2004. How to Develop New Approaches to RFM Segmentation. Journal of Targeting Measurement and Analysis for Marketing. Volume 13. Issue 1. Pages 50–60. ISSN 0967-3237. {https://doi.org/10.1057/palgrave.jt.5740131}.
  • Yim et al. (2014) Yim, M. Y.-C., Yoo, S.-C., Sauer, P. L., Seo, J. H. 2014. Hedonic Shopping Motivation and Co-Shopper Influence on Utilitarian Grocery Shopping in Superstores. Journal of the Academy of Marketing Science. Volume 42. Issue 5. Pages 528–544. ISSN 0092-0703. {https://doi.org/10.1007/s11747-013-0357-2}.
  • Zhang et al. (2007) Zhang, Y., Jiao, J., Ma, Y. 2007. Market Segmentation for Product Family Positioning Based on Fuzzy Clustering. Journal of Engineering Design. Volume 18. Issue 3. Pages 227–241. ISSN 0954-4828. {https://doi.org/10.1080/09544820600752781}.