Session-based recommendation with hypergraph convolutional networks and sequential information embeddings

Abstract

Session-based recommendation focuses on predicting the next item that an anonymous user is most likely to click. Due to its privacy-protecting ability, it is receiving increasing attention from researchers in recent years. The existing studies typically focus on sequence or graph structure learning. However, they ignore the consistency and sequential dependence relationships that are widely existed between items in real-world scenarios. To address this problem, we present a novel method named HyperS2Rec. Specifically, we propose to learn two kinds of item embeddings with hypergraph convolutional network and gated recurrent unit, respectively, to account for both consistency-awareness and sequential dependence-awareness. Then the attention mechanism is designed to flexibly combine the above both embeddings. Finally, the reversed position and the soft attention mechanism are utilized to obtain session representations. To verify the effectiveness of the proposed HyperS2Rec, we conduct experiments on three real-world datasets. The results prove that the proposed HyperS2Rec significantly outperforms state-of-the-art methods. The source code of our proposed model is available at https://github.com/ZZY-GraphMiningLab/HyperS2Rec.