Contextual attention model for social recommendation

Abstract

Recently, with the emergence of a large number of social platforms, more and more works have been explored for social recommendation. On a social platform, social scientists converged that there exists social influence among users. Thus, accurately modeling the social influence could alleviate the data sparsity issue in Collaborative Filtering (CF). Most of the methods simply define the social influence with the normalized constant weights. However, this is not accuracy enough, which requires more reliable modeling. Besides, many studies have adopted neural network with CF in various recommendation tasks due to the effective ability of neural network for representation. In this paper, we attempt to apply attention mechanism based neural network structure for social recommendation. Specifically, social attention can weigh the contribution of social influence in the form of scores from each neighbor, and then generates each user’s social context. Finally, extensive experimental results confirm the feasibility and effectiveness of our proposed model.