Recommendation System Based On Graph Spatial-temporal Network

With the rapid development of the Internet,people cannot get the information they want from the huge amount of data,and recommendation systems can solve this problem,so more and more people in industry and academia are studying it deeply.The graph structure in the data information can represent a class of items and the diverse relationships among them.Since most of the data information in the recommendation system has a graph structure,the graph neural network can understand the users’ preferences and needs from various data information.Therefore,a key research direction of the recommendation system is the recommendation system based on graph neural networks.According to the existing graph spatiotemporal network in the graph neural network,the new algorithm is proposed to enhance the accuracy of product recommendation on the e-commerce platform.The core of this study is as follows:(1)First of all,the recommendation system for graph neural networks does not fully explore the data information in the time series,and will give the same weight to different neighbor items to represent the user’s preferences.In this paper,we propose a recommendation system approach based on graph spatiotemporal networks.The core step of this method is to first build the embedded representation matrix of the initial user and the project;then aggregate the neighbor data information,which add sequential attention weights to represent different attention relationships;then obtain the embedded representation of the neighbor aggregation layer output and stack multiple propagation layers for propagate;then obtain the embedded representation of the propagation layer,obtain the embedded vector and calculate the inner product;finally evaluate the inner product result,and recommend the item to the user.(2)Secondly,The one-way data information structure limits the important information hidden in the sequence of user behavior.To address these limitations,this paper proposes a sequential recommendation model which uses two-directional self-attention for user behavior sequences.To effectively train the model effectively,randomly hidden items in the sequence are first predicted by combining the context of the project,and finally the complete target will be supplemented for sequence recommendation.In this way,we allow each item in the user’s history to combine information on the left and right sides to make recommendations.To sum up,this paper puts forward some solutions and ideas for the problems and difficulties in the recommendation system based on graph neural network,and carries out experimental verification,which provides some reference value for some meaningful research directions of the recommendation system.