Enhancing Session-Based Recommendations by Fusing Candidate Items

Session-based recommendations are used to convert complex items by using the graph neural network, where this also involves combining session-level and global-level information to discover user preferences. However, this ap-proach en-counters certain problems. A user with extensive interests should be offered more than one recommendation of candidate items. We propose a neural network-based model to fuse candidate items based on this premise. We first use a graph neural network to acquire session-level and global-level information, and then use an attention mechanism to obtain a representation of candidate items recommended to the user. Finally, we integrate the candidate-level, glob-al-level, and session-level information to acquire rich information on the items in the given session. Extensive tests on three empirically ac-quired datasets showed that our model is superior to baseline models in most cases.