Font Size: a A A

Research And Application Of Recommendation Algorithm Based On Knowledge Graph

Posted on:2024-08-04Degree:MasterType:Thesis
Country:ChinaCandidate:Z Y WangFull Text:PDF
GTID:2568306914469284Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Due to the explosive growth of Internet data resources,the ensuing information overload problem has become increasingly prominent.The recommendation system can accurately recommend the content of interest to users by predicting user preferences,and can help users quickly filter out effective information from huge amount of data,so it has been increasingly used for movies,music,short videos and so on.The data sparsity and cold-start problems prevalent in traditional recommendation algorithms leave much to be desired in terms of making reasonable recommendations.The rich entity and relationship information contained in knowledge graphs can deliver effective performance enhancements for recommendation systems,and solving the problems of traditional algorithms through knowledge graphs is gradually attracting a lot of focus from researchers.The Knowledge Graph Attention Network(KGAT)recommendation model combines user interaction history and knowledge graphs,which can explicitly model high-order relationships in graphs in an end-to-end manner,and use attention mechanisms to distinguish neighbors The importance of nodes is used to optimize the propagation embedding between nodes,and finally achieve effective recommendation by aggregating the embedded representations of users and items.Since the model propagates information across the entire knowledge graph,it is easy to introduce irrelevant entities,resulting in the final user and item representations being susceptible to noise.Aiming at this problem,the main work of this thesis includes the following:(1)To address the problems of KGAT recommendation model,by improving the attention score strategy and adding information filtering layers,this thesis proposes a Knowledge Graph Attention Network with Information Filtering(KGAT-IF).The improvement of the attention scoring strategy aims to take into account both the differences between nodes and the vector offset to more accurately reflect the distance and relationship between nodes,so that more useful information can be transmitted between similar entities;the research on the impact of noise aims to filter the nodes with low similarity by adding an information filtering layer to reduce the noise influence in the information dissemination process and optimize the node embedding.Experiments are conducted on two publicly available datasets,Amazon-Book and Last-FM,and the experimental results show that the improved KGAT improves on both recall and NDCG evaluation metrics,including 1.54% and 1.68% on Amazon-Book and 1.03% and 1.96%on Last-FM effectively improving the recommendation effect.(2)Through the requirement analysis and system design of the personalized movie recommendation system,a knowledge graph attention network recommendation model based on information filtering with a threshold value of 0.4 is implemented,and the functions of movie recommendation,movie details and movie collection are realized,and the testing of each part of the system is completed.
Keywords/Search Tags:recommendation system, knowledge graph, graph neural network, attention mechanism, information filtering
PDF Full Text Request
Related items