Dynamic global structure enhanced multi-channel graph neural network for session-based recommendation.

Session-based recommendation is a challenging task, which aims at making recommendation for anonymous users based on in-session data, i.e. short-term interaction data. Most session-based recommendation methods only model user's preferences with the current session sequence, which ignore rich information from a global perspective. Meanwhile, previous works usually apply GNN to capture the transformation relationship between items, however the graph used in GNN is built through a static mode, which may introduce noise to the graph structure if user's preferences shift. In this paper, we propose a novel method called Dynamic Global Structure Enhanced Multi-channel Graph Neural Network (DGS-MGNN) to learn accurate representations of items from multiple perspectives. In DGS-MGNN, we propose a novel GNN model named Multi-channel Graph Neural Network to generate the local, global and consensus graphs dynamically and learn more informative representations of items based on the corresponding graph. Meanwhile, in order to reduce the noise information within sessions, we utilize the graph structure to assist the attention mechanism to filter noisy information within each session, so as to generate an accurate intention representation for the user. Finally, combined with a repeat and explore module, a more accurate prediction probability distribution is generated. We conduct extensive experiments on three widely used datasets, and the results demonstrate that DGS-MGNN is consistently superior to the state-of-the-art baseline models. [ABSTRACT FROM AUTHOR]