Research On Recommendation Algorithms Based On Graph Data

The development of the Internet has enriched the way people obtain information,but at the same time,a series of problems such as information overload have arisen.How to extract the content that users intend to know from massive information has become a hot research topic.Recommendation algorithms can effectively solve existing problems and has been widely used in many scenarios,but the recommendation algorithm also has certain problems,that is,the cold start problem that new users will face when they log in to the old platform or when the new software is just put into use;while people enjoy the convenience of the recommendation algorithm,they hope that the platform can give a reasonable explanation for the recommendation results,that is,to improve the interpretability of the recommendation algorithm;the recommendation performance of the recommendation algorithm in the sparse environment needs to be improved,etc.Based on graph data,this thesis improves the recommendation algorithm based on graph neural network.The main contributions are as follows:(1)This thesis proposes the LNGCF-B(Light Neural Graph Collaborative Filtering with user Basic information),which is based on the collaborative filtering algorithm,considers user’s attribute information,and uses graph convolutional network to represent the relationship between users and items.In order to study the recommendation performance of the algorithm in different application scenarios,select two datasets belonging to different network models and visualized for analysis.Through experiments to analyze the recommendation performance of the algorithm in this thesis under different network models,and compared with the benchmark algorithm,the experiments show that for different application scenarios,the optimal parameters of the algorithm are different,compared with the benchmark algorithm,the algorithm in this thesis has improved in Recall and Precision.(2)In order to better integrate auxiliary information and alleviate the problems existing in collaborative filtering algorithm,we propose the SKGPN(Recommendation Algorithm Based on Strengthening Knowledge Graph and Integrating User Interest Preferences),which is based on the enhanced knowledge graph and integrates the interests of users.The algorithm extracts the interests of users,and expresses the relationship between users and items in a more detailed manner.While considering the semantic information in the knowledge graph,it integrates the structural information between entity nodes,finally,use MLP multilayer perceptron for prediction.Several comparative experiments have been done on datasets with different degrees of sparseness,and it is found that the algorithm in this thesis is more suitable for sparse scenarios than the benchmark algorithm.