Social Recommendation Based on Users’ Attention and Preference

Jiawei Chen,Can Wang,Qihao Shi,Yan Feng,Chun Chen

DOI: https://doi.org/10.1016/j.neucom.2019.02.045

IF: 6

2019-01-01

Neurocomputing

Abstract:Attention is the behavioral and cognitive process of selectively concentrating on small fraction of information while ignoring other perceivable information. Thus, user’s attention will influence his decision on the consumption and consequently the performance of recommender systems. When presented a long list of recommendations, users will focus more on the items that draw their attention and skip others even if they suit their tastes. However, existing works in social recommendation mainly generate recommendations based on users’ preference and overlook the importance of users’ attention. To overcome this limitation, in this paper we propose a probabilistic model HTPF that considers both users’ attention and preference in social recommendation. Besides, by analyzing four real-world social recommendation datasets, we observe that social relations have more explicit impact on users’ attention than on their item rating values. That is, user’s attention on items are easily affected by his social relations while his rating values seem somehow indifferent to social influence. It will be more effective to use social network information to infer users’ attention rather than users’ preference. Thus, different from existing methods that model the social influence on user’s rating values, HTPF focuses on exploiting social influence on user’s attention. Further, a scalable variational inference algorithm has been developed for our model to predict how the specific items will draw users’ attention and how the items will suit users’ tastes. Then HTPF generates the recommendation results based on the weighted combination of these two aspects. Extensive experiments conducted on large real-world datasets validate the superiority of our method over state-of-the-art methods.