Research On Graph Neural Network Recommendation Algorithm Based On Spatio-temporal Information

The recommendation system is designed to filter out the information that matches the user from a large amount of overloaded information.In real life,this is of great significance to both the user and the merchant.Because the graph neural network can use the spatial structure to discover the connection between users and items,it continues to flourish in the research of recommendation.However,there are still some problems in the process of modeling user sequences in previous studies:(1)The field of view is limited to the perspective of spatial structure,ignoring the importance of temporal information in capturing user interest task.(2)As user sequence length grows over time,how to fully consider the impact of long-term preferences in the modeling process.To this end,this paper proposes a graph neural network recommendation model based on spatio-temporal information.The main research work and innovations are as follows:(1)This paper designs an advanced spatio-temporal information aggregation algorithm TAP(Temporal and Positional aggregation algorithm,TAP).It does not only aggregate neighbor information on the spatial structure of the graph,but synchronously aggregates the embedded representation information from the spatial structure and the time sequence of user-item interaction,and then fully considers the time information in the recommendation,which promotes the original model from space to space aggregation to the development of spatio-temporal aggregation.TAP utilizes scaled dot-product attention to capture the temporal order of one-hop neighbors,and employs a multi-head attention mechanism for joint attention representations for different latent subspaces.Finally,TAP is indiscriminately added to the recommendation based on the graph neural network to form a new model TAP-GNN to learn the embedded representation of users and items.(2)This paper proposes a recommendation model based on memory augmented spatio-temporal perception information MTAP-GNN(Memory enhancement Temporal and Positional Graph Neural Network).It adds an external memory network,and fully considers the user’s short-term preference and long-term memory.In order to effectively integrate users’ multi-angle interests,this paper introduces a gated learning network into the model framework,and adaptively combines these two embedded representations.to learn embedding representations of users and items.(3)This paper conducts experiments on three real data sets.The experimental results show that the TAP-GNN model and the MTAP-GNN model have excellent performance compared with the previous classic recommendation models in their respective fields.