BiG: Bidirectional Graph Convolutional Networks for Social and Sequential Recommendation

Graph convolutional networks have obviously influenced recommender systems. However, we argue that previous research has limitations: 1) unidirectional approaches are not enough to explicitly capture the representation learning, 2) social influence is ignored to extract the user preference drifts, and 3) simple and crude fusion methods are not enough to seamlessly amalgamate the diverse behavioral perspective. To address these limitations, this research proposes a novelty of Bidirectional Graph convolutional networks for social and sequential recommendation (BiG). A novel graph convolutional networks are modified into bidirectional graphs to explicitly learn the representation of dynamic user preference at social level and sequential level. A novel Bidirectional Gate (BiGate) is designed to amalgamate the diverse preferences of social and sequential influences. Empirical experiment illustrates that BiG outperforms state-of-the-art methods.