关键词: 会话推荐, 全局图, 会话图, 图神经网络, 邻域信息

Abstract: To solve the problem that representation of interest preferences based on the recurrent neural network is incomplete and inaccurate in session recommendation, a session recommendation based on graph model and attention model（SR-GM-AM） method was proposed. Firstly, global graph model and session graph model were used to obtain neighborhood information and session information respectively, the item graph features were extracted through the graph neural network, global item representation and session item representation obtain two-level representations, which were combined with the graph embedding. Then, soft attention was used to fuse graph embedding and reverse position embedding, target attention activated the relevance of the target items, session embedding was generated through linear transformation. Finally, the recommended list of the N items for the next click was outputted through the prediction layer. The comparison experiments were carried out in the public data sets. Experimental results show that, P@20 reaches 72.41% and MRR@20 reaches 35.34%, compared with the optimal Lossless Edge-order preserving aggregation and Shortcut graph attention for Session-based Recommendation(LESSR) method, it is improved by 0.5 percentage point and 2 percentage point respectively, verifying the effectiveness of the proposed method.

Key words: session recommendation, global graph, session graph, graph neural network, neighborhood information