Temporal-aware and multifaceted social contexts modeling for social recommendation

The social recommendation has utilized ego network in a static manner, assuming that the way of social influence impacts users’ preferences is simplex and constant over time. Such assumptions hinder the effectiveness of social recommendation since the online social contexts are complicated and friends’ tastes may change across items consumption. To address this research gap, we propose a novel social recommendation model that focuses on the effect of different social contexts and the dynamics of social influence. Specifically, we first leverage a community detection algorithm to divide online users into various communities, named community-based social contexts. Based on the community structure, we introduce a hierarchical attention mechanism to evaluate the influence of multiple community-based social contexts. Secondly, we choose the directly connected neighbors, namely the ego-network based social contexts. In such social contexts, the model can capture the user’s preferences that are influenced by their friends’ dynamic behaviors. Then, the model could have a comprehensive representation of users’ current preferences by merging the impact of two types of social contexts. Extensive experiments on three real-world datasets demonstrate that our proposed model outperforms the state-of-the-art social recommendation models. The implementation of our model and the datasets are available at https://github.com/socialsnail/TSC.