Multi-context aware user–item embedding for recommendation

Abstract

Real recommender systems usually contain various auxiliary information. Some of the most recent works make meaningful exploration of incorporating auxiliary information into the representation model for competitive recommendation. However, learning user and item representations still faces two challenges: (1) existing works do not well address the problem of integrating multi-type auxiliary information; (2) learning representations for inactive users is still challenging due to the high sparsity of explicit user–item associations. In order to tackle these problems, in this paper, the attributed heterogeneous network and bipartite interaction network are employed to incorporate various auxiliary information and user–item associations. A joint objective function and an efficient algorithm are devised for the representation learning. Experimental results show that the proposed algorithm has significant advantages over the state-of-the-art baselines. What is remarkable is that our proposed method is demonstrated to be especially useful for dealing with low-active users in the system.