Research On Reinforcement Session Recommendation Algorithm Based On Hypergraph

In order to protect user information,there are more and more anonymous user interaction sessions in common session recommendation scenarios,and session recommendation systems mainly based on anonymous sessions are gradually developing.However,the current supervised learning modeling method commonly used in the field of session recommendation has a sub-optimal recommendation problem,and the research on using supervised reinforcement learning to solve this sub-optimal recommendation problem is relatively few.To enhancing the recommendation performance of session recommendation systems in anonymous environments,this paper utilizes a supervised reinforcement learning recommendation framework to incorporate complex dependencies of items in different sessions into the state representation of reinforcement learning from both global session hypergraph and local session graph.Based on this,a hypergraph based method for obtaining nearest neighbor sessions is studied to generate better recommendation results.Firstly,in order to address the suboptimal recommendation problem in supervised session recommendation algorithms and construct a better state representation method for reinforcement learning algorithms,this paper proposes a supervised reinforcement session recommender model based on global hypergraph and local session graph(HG-SRL).In this model,this article proposes for the first time a state representation method based on global hypergraphs and local session graphs to fill the gap in the graph state construction method of supervised reinforcement learning recommendation algorithms.This method obtains high-order item relationships contained in the global session hypergraph through a hypergraph convolutional network,and mines pairwise item dependencies contained in the local session graph through a gated graph neural network.Secondly,in order to fully utilize session information at different levels,this paper proposes a self matching attention mechanism to fuse session information at both global and local levels.By cross computing session item features from both perspectives,the feature information of different levels contained in them is embedded into the final graph state representation.Thirdly,in order to make the state representation of reinforcement learning more comprehensive,this paper proposes a method for mining the nearest neighbor sessions of each anonymous session based on the global session hypergraph.By obtaining item adjacency relationships on the global session hypergraph to supplement the nearest neighbor information of real-time sessions,the state contains more nearest neighbor item features.Finally,effectiveness experiments and comparative experiments conducted on three real datasets show that the proposed enhanced session recommendation model based on global session hypergraph and local session graph can effectively improve recommendation performance on anonymous session data.