Research And Application Of Recommendation Algorithm Based On Graph Neural Network

With the rapid development of the Internet,the problem of information overload is becoming more and more prominent.The recommendation system is a solution proposed to solve the information overload.It selects the part that meets the target user's interest and preferences from the massive amount of information.The rating prediction task is one of the most important tasks in the recommendation system.Only by accurately predicting the user's rating on unknown items can a better recommendation effect be achieved.In order to solve the tasks in the recommendation system,people have proposed many kinds of recommendation algorithms.These algorithms use different data or analyze the data in different ways,and they are divided into different types.The work of this paper revolves around the rating prediction task in the recommendation system.Based on the idea of collaborative filtering,the historical interaction data of users and items is mainly used to solve the task of rating prediction.Different from the traditional collaborative filtering algorithm that expresses user-item interaction data in the form of a rating matrix,this paper converts user-item interaction data into a heterogeneous graph,and regards users and items as two types of nodes on the graph,regards the user-item rating as an edge connecting the two types of nodes;so the rating prediction task in the recommendation algorithm is transformed into a link prediction task on the graph data.The Graph Neural Networks is a general term for a type of neural network dedicated to processing graph data.This article builds recommendation algorithm model based on related technologies.This article proposes some modifications to the GC-MC algorithm model,designs a recommendation algorithm model based on Graph Convolutional Networks.In the model,the graph convolutional operator is used to aggregate the features of its neighbor nodes for each node on the graph to generate latent features to represent each node,and the latent factor of users and items are used to make rating predictions.Then,an improved model by introducing social information is proposed.In the improved model,the social relationship between users is introduced,and the social relationship is combined with the latent factor used to represent the user node,so that the latent factor can more accurately represent the user.Finally,this article designs and implements a singer recommendation system.The front-end part of the system uses the Vue framework and Element UI component library,the back-end part uses the FastApi framework,and the recommendation module of the system is combined with the improved recommendation algorithm model of this article.