Research On Graph Neural Network For Sequential Recommendation Based On Contrastive Learning

With the advent of the Internet era,people are facing the growth of massive information,which leads to problems such as information overload.In order to help users better filter information,recommendation systems have emerged,which can help users get personalized content in the information ocean.Sequential recommendation is one of the important scenarios,because human behavior is dynamic and continuous,how to judge the user’s next action based on their historical behavior sequence is an important task of sequential recommendation.And user behavior sequences often have complexity and diversity,different user sequences also have relevance,how to accurately capture the dynamic preferences of user sequences and complex information between sequences is a difficult problem faced by recommendation systems.This paper mainly focuses on the research of graph neural network sequential recommendation method based on contrastive learning.At the same time,this article explores the modeling role of graph neural networks in sequential recommendation and the application of contrastive learning methods in order to achieve better recommendation results.The main research contents of this paper are:(1)Aiming at the problem that the traditional single-sequence model can’t obtain dynamic collaboration information across sequences and sequence sparsity,this paper proposes a dynamic graph neural network sequential recommendation method based on contrastive learning.This method integrates sequence information into a dynamic graph neural network,learns and mines complex relationships between sequences through multi-layer networks to obtain user preferences,and generates multiple views through contrastive learning to increase supervision signals.Through joint learning to obtain more accurate user node representation,perform next item recommendation.Finally,experiments are conducted to verify the effectiveness of the proposed method.(2)Aiming at the problem of not being able to obtain item side information and lacking supervisory signals,this paper proposes a graph neural network sequence recommendation method based on bidirectional contrastive learning.This method converts sequences into graph structures,and then samples subgraphs from two perspectives of user side and item side for modeling and learning at the same time.Through graph neural network,it obtains user node representation and item node representation respectively,performs contrastive recommendation from both ends,and continuously optimizes user side rating matrix to improve recommendation effect.Finally,experiments are conducted to verify the effectiveness of the proposed method.(3)Based on the two proposed methods in this paper,this article implements an ecommerce recommendation system that models user historical behavior based on graph neural networks and contrastive learning,providing personalized product recommendations to users.