Research On Personalized Recommendation Algorithm Based On Knowledge Graph And Deep Learning

In recent years,recommendation technology has been widely used in e-commerce,social media and search engines to provide users with personalized information services.However,in the face of complex application scenarios and business environments,personalized recommendation technology has the following problems: 1)The sparse historical interaction data of users leads to low accuracy of personalized recommendation results;2)The personalized recommendation model is difficult to capture the dynamic changes of user interests.Affect the accuracy of recommendations.Aiming at the above two problems,this paper proposes a personalized recommendation model-CSGAT based on the knowledge graph,using graph convolutional neural network and long-short-term memory neural network.The main research contents are as follows:(1)Aiming at the problem of low accuracy of recommendation results caused by sparse data in recommendation technology.This paper integrates users,items and knowledge graphs to build a unified knowledge graph structure,uses knowledge representation learning and graph attention mechanisms in the knowledge graph,aggregates the rich entity semantic knowledge to users and project nodes,it can make up for the lack of user and item's own characteristic information under the sparse data situation,and improve the accuracy of recommendation.(2)Aiming at the problem that it is difficult to capture the dynamic changes of user interest in the existing recommendation technology.The model proposed in this paper first uses the graph convolutional neural network to aggregate the semantic knowledge information of the knowledge graph to the project node,and then uses the long and short-term memory neural network to capture the time characteristics of the project node and aggregate it to the user node,thereby dynamically capturing the user's interest changes.Accurately characterize the user's interest.(3)The experimental part is compared with baseline methods such as CFKG,Lib FM,Ripple Net,MKR,KGCN,etc,in the CTR prediction and Top-K recommendation scenarios.The results show that the recommended accuracy of the CSGAT model proposed in this article is better than the baseline method.Recommendation quality and performance in sparse scenarios have been improved.At the same time,multiple sets of comparative experiments are set up to prove that the model in this paper can capture the dynamic changes of user interests and provide dynamic personalized recommendation results.Apart from this,in order to further upgrade the performance,this paper improves the sample sampling process,and dynamically generates training samples based on the historical data of user interaction and the knowledge information in the knowledge graph,and the experiment proves that this method can effectively improve the training speed of the recommendation model.