Learning Product Embedding from Multi-relational User Behavior.

Network embedding is a very important method to learn low-dimensional representations of vertexes in networks, which is quite useful in many tasks such as label classification and visualization. However, most existing network embedding methods can only learning embedding from single relational network, which only contains one type of edge relationship between two nodes. However, in real world, especially in product network, many information is presented in multi-relational network. Based on user behavior, edges in product network have many types: "co-purchasing", "co-viewing", "view after purchasing" and so on. Therefore, we propose a novel network embedding method aiming to embed multi-relational product network into a low-dimensional vector space. The results show that our method leads to better performance on label classification and visualization tasks in product network.