Research Of Session-based Recommendation With Graph Neural Network

Recommendation technology can recommend products that users are interested in,and has been applied to all aspects of the Internet.In a general recommendation system,recommendation decisions are usually made based on the user's complete personal information and historical behavior.However,in some cases,the user's login and access are anonymous,their personal information and configuration files are often not directly available,and the user's interests are dynamically changing and have strong immediacy.Therefore,it is necessary to consider the problems which are only based on the user's current ongoing session(also known as an anonymous session)and makes a recommendation decision,also known as session-based recommendation.Some early studies regarded sessions as linear sequences,and used Markov chains,recurrent neural networks and other methods to model sessions.The effects of these methods are acceptable,but when they model and represent the complex relationship between the sessions and multiple items,they have caused a loss of information and cannot accurately express the meaning of the data itself;and they have not taken into account the session itself.The problem of temporal and spatial adaptability changes,and there is no effective mining of knowledge from historical data.In order to solve the above problems,this thisis proposes a new session-based recommendation model with graph neural network.In the proposed method,this thisis firstly completes a complete session-based recommendation model based on a gated graph neural network and a self-attention network;in order to solve the problem of information loss in the process of encoding a session sequence into a session graph,a new lossless session is proposed.A graph coding method is used to build session graphs to improve graph neural networks;in order to solve the problem of adapting to temporal changes of sessions,a new meta-learning method is proposed to strengthen the learning ability of the model.Finally,this article proposes a brand new session-based recommendation model.This thesis conducts multiple experiments on two benchmark datasets.By comparing existing methods based on traditional machine learning and methods with neural networks,two recommended evaluation indicators are used to verify the performance of the proposed model.The experimental results show that our method in this thesis achieves the best results in evaluation indicators,which is significantly better than the latest methods in the current session-based recommendation field.