Personalized Tag Recommender System Based On Graph Convolutional Neural Network

With the rapid development of the Internet,the functions of Web2.0 applications have become more and more rich,and many user generating content systems give tagging permissions to users.Tags not only enrich data resources in the internet,but also can be applied to many scenarios such as information retrieval and recommender systems.However,the behavior of social tagging is entirely determined by users,which leads to a large number of redundant and irregular tags.Therefore,it is of great practical significance to conduct research on the technique of tag recommendation.On the one hand,tag recommendation is helpful to remove redundant and non-standard tags,and could create better tag data resources for application systems;on the other hand,tag recommendation is also a service provided for users by the application system.So a good recommendation service can enhance the user experience.However,currently most tag recommendations rely on traditional algorithms such as content-based.In addition,many algorithms only consider the relationship between tags and resources and ignore the importance of users,so the recommendation performance is not good.Due to its excellent performance on graph representation tasks,graph neural networks have become a current research hotspot,and the combination of graph neural networks and recommender systems may further improve recommendation performance.Therefore,this thesis designs a personalized tag recommender system based on graph convolutional neural network.With the help of the representation learning ability of graph convolutional network,the designed algorithm can be used to recommend tags for resources published by users.The main work of this thesis includes the following:(1)A personalized tag recommendation algorithm based on graph convolutional neural network is proposed.The design idea and implementation method of the algorithm are illustrated in detail,which mainly includes three parts:data preprocessing and feature extraction,representation learning for users and tags,and personalized tag recommendation.(2)A personalized tag recommender system is designed and implemented.Firstly,the system's functional requirements were clarified through the system's requirements analysis.Secondly,this thesis modularize the functional requirements through system design,including system architecture design,functional module design,and database design.Finally,the related environment of the system implementation and the implementation method of each function module are stated,and the implementation effect of the system is displayed in the end.(3)An experimental verification analysis of the algorithm in this thesis is conducted.The effect of algorithm parameters on the performance of recommendation is tested,and the effect of attention mechanism on the performance of recommendation is tested.In addition,the thesis conducted comparative experiment with common tag recommendation algorithm.From the experimental results,we could conclude that the proposed algorithm in this thesis has a certain improvement in the recommendation precision and recall rate compared with the common tag recommendation algorithm,which proves the effectiveness of the proposed algorithm.