Neighbor-items Aware Graph Neural Networks for Session based Recommendation for large rotating units

A lot of results have been achieved so far for session-based graph neural network recommendations, which aim to predict user behavior based on anonymous sequences of user actions. However, a large number of graph neural network-based session recommendations focus on the current session, which is short in length and contains limited information in most cases. Therefore, in this paper, we propose a novel graph neural network recommendation model based on neighbor item awareness. The model designs a session-aware encoder that efficiently aggregates neighbor item information with the help of global session information, and achieves session information enhancement while reducing the introduction of noise. Specifically, NA-GNN constructs a global session graph and a current session graph to model the influence of neighboring items on items and sessions: (1) Global session graph, by creating links to related items in all sessions and constructing root-mean-square-valued session perceptrons to explore the influence of neighboring sessions on items, and finally fusing out session representations of neighboring item features. (2) Current session graph, exploring item embedding by modeling pairwise item transitions in the current session. And, in this paper, we fuse the session representation with neighboring item information and the current session representation through an attention mechanism. Extensive experiments on two real-world datasets show that our approach consistently outperforms state-of-the-art methods.