Session-based recommendation (SBR) aims to predict a user's next clicked item based on an anonymous yet short interaction sequence. Previous SBR models, which rely only on the limited short-term transition information without utilizing extra valuable knowledge, have suffered a lot from the problem of data sparsity. This paper proposes a novel mirror graph enhanced neural model for sessionbased recommendation (MGS), to exploit item attribute information over item embedding vectors for more

more »

... te preference estimation. Specifically, MGS utilizes two kinds of graphs to learn item representations. One is a session graph generated from the user interaction sequence describing users' preference based on transition patterns. Another is a mirror graph built by an attribute-aware module that selects the most attribute-representative information for each session item by integrating items' attribute information. We applied an iterative dual refinement mechanism to propagate information between the session and mirror graphs. To further guide the training process of the attribute-aware module, we also introduce a contrastive learning strategy that compares two mirror graphs generated for the same session by randomly sampling the attribute-same neighbors. Experiments on three real-world datasets exhibit that the performance of MGS surpasses many state-of-theart models.