Research And Application Of Graph Neural Network Recommendation System Based On Knowledge Graph

In the era of big data,recommendation systems can effectively solve the problem of information overload.As a heterogeneous information network with rich information,the knowledge graph is introduced into the recommendation system,which can help generate better recommendations.However,with the continuous enrichment of available information,the construction of the knowledge graph and its application in the recommendation system have become a key challenge.In construction,it is necessary to solve the difficulties of heterogeneous data extraction,multi-source information fusion and knowledge storage;in application,the traditional path-based and embedding-based recommendation methods have limited utilization of the information in the knowledge graph.Therefore,studying how to effectively construct and apply knowledge graphs is a subject of important practical significance.This paper proposes a domain knowledge graph construction method based on heterogeneous information fusion,which constructs domain knowledge graphs through information extraction,knowledge fusion and other technologies and applies them to recommendations.And further use the graph neural network to conduct in-depth information mining on the knowledge graph,and design a graph neural network model that combines deep-domain information exploration and wide-area information exploration,which verifies the improvement of the recommendation effect on the real data set,and Explore the application of domain knowledge graphs in the automotive field.The main work of this paper is as follows:First,analyze the current research and application status of knowledge graphs and graph neural networks in the direction of recommendation systems,conduct in-depth analysis and reflections on the recommendation methods currently in research,explore the possibilities and technical solutions applied to car recommendation,and make improvements on this basis and practice.Second,on the basis of the existing knowledge map construction method,a domain knowledge map construction method that integrates multi-source heterogeneous information is proposed,and targeted solutions are proposed to solve the difficulties of information extraction and knowledge fusion,and the actual application scenarios are collected.Experiments on real data,through the quality evaluation of the completed domain knowledge map,verify the effectiveness of the method.Third,aiming at the problem of insufficient utilization of knowledge map information by existing embedding-based methods and path-based methods,a recommendation model based on graph neural network,WDEG model,is proposed.First,the graph embedding technology is used to pre-train the knowledge graph to obtain the network Using graph convolution technology to conduct in-depth semantic information mining on the knowledge graph,and finally verify the performance of the model through multiple experiments.Last,on the basis of the above-mentioned practice,the Django framework is used to build a server to realize a car recommendation system,which has core functions such as login and registration,popular recommendation,personalized recommendation,and conditional car selection.