LATTE: Application Oriented Network Embedding
share
In recent years, many research works propose to embed the networked data into a low-dimensional feature space. With the embedding feature vectors, the original network structure can be effectively reconstructed, classic learning algorithms can be applied directly, and more importantly the learned embedding representations can also be widely used in external applications. However, due to the detachment of embedding process with external tasks, the learned embedding results by most existing embedding models can be ine ective for application tasks with speci c objectives, e.g., community detection vs information di usion. In addition, the networked data has become more and more complicated nowadays, which can involve both heterogeneous structures and diverse a ributes, and few existing homogeneous network embedding models can handle them well. In this paper, we will study the application oriented heterogeneous network embedding problem. Signi cantly di erent from the existing works, besides the network structure preservation, the problem should also incorporate the objectives of external application in the objective function. To resolve the problem, in this paper, we propose a novel network embedding framework, namely the "appLicAtion orienTed neTwork Embedding" (Latte) model. In Latte, we introduce a new concept called "attributed heterogeneous social network" to model the diverse structure and attribute information available in the networks. Meanwhile, the heterogeneous network structure can be applied to compute the node "diffusive proximity" score, which capture both local and global network structures. Furthermore, Latte learns the network representation feature vectors by extending the autoencoder model model to the heterogeneous network scenario, which can also effectively unite the objectives of network embedding and external application tasks.
READ FULL TEXT
