Research On Recommendation Model Based On Knowledge Graph And Neural Network

interpretability,and diversity of recommendations.Combining knowledge graphs and graph neural networks to improve recommended performance has become a hot topic in current research.Although the recommended performance of KGCN,the most advanced recommendation model combining knowledge graph and graph neural network,has achieved good results,the user side has not yet been modeled,which limits the further improvement of recommendation effectiveness,and does not make full use of the highorder connectivity between nodes in the knowledge graph,resulting in poor interpretability of recommended results.Therefore,this thesis mainly considers the improvement of the existing algorithm from the aspects of recommended validity and interpretability,and proposes KBGNN model and PGPR-ER model.The main work of this article includes :(1)By improving the KGCN model,a recommended model KBGNN of bilinear graph neural network based on knowledge perception is presented.On the basis of KGCN,the model adds the source of information for the model by introducing user comments as background knowledge to describe the user,uses max pooling aggregator to aggregate the most representative information in the neighborhood,and uses bilinear graph neural networks instead of traditional graph neural networks for graph convolution operations to effectively improve the effectiveness of recommendations.(2)An interpretable recommendation model PGPR-ER of graph neural network based on knowledge perception is proposed.The model can use the additional knowledge about the project in the knowledge graph,introduce the strategy-guided path inference method to find different recommended reasoning paths,save the path,and interpret the recommended results through radar charts or stylistic interpretations,so that the recommendation model has interpretability.(3)The above two models are implemented and applied.Mainly using the GPU version of the graphics processing tool tensorflow,with scipy,cycler,numpy and pandas and other tools for data processing,improving the efficiency of graph operations.And use the Twisted framework to write tcp services and Http services to implement the server side of the recommendation system,which provides convenience for customers to use.Top-N recommendations and scoring predictions are made on the recommendation models KBGNN and PGPR-ER on three datasets,Apps for Android,Movies and TV,and Movie Lens 20 M,respectively.The experimental results show that KBGNN is better than KGCN and baseline model in terms of accuracy,recall rate,F1 measure,mean square error,root mean square error and average absolute error;PGPR-ER is comparable to KGCN in these six indicators;the results of interpretable exploration experiments show that PGPR-ER can be interpreted for the recommended results;and the results in time performance comparison experiments show that PGPR-ER is greater than KBGNN model in terms of time consumption and computational resource consumption.