Font Size: a A A

Research On Recommendation Algorithm Based On Knowledge Graph

Posted on:2023-07-18Degree:MasterType:Thesis
Country:ChinaCandidate:K N ZuoFull Text:PDF
GTID:2558307070984009Subject:Engineering
Abstract/Summary:PDF Full Text Request
As a structured knowledge base containing rich information,knowledge graph can describe entity characteristics and relationships in reality,and provide rich semantic information as auxiliary information.In this Internet with serious information overload,in order to improve the existing personalized recommendation model,the combination of knowledge graph and recommendation system has become one of the current research hotspots.This thesis studies entity and relation representation in knowledge graph based on graph convolutional network,and integrates attention mechanism,proposes the bilinear network recommendation algorithm based on knowledge graph and knowledge graph completion method based on graph attention network,achieves effective representation of entity information in knowledge graph,enhance the performance of the recommendation effect,It solves the problem of poor recommendation effect in the case of sparse data or incomplete entity information.The main research work of this thesis is as follows:(1)Bilinear network recommendation model based on knowledge graph.In view of the lack of feature interaction information between nodes in existing recommendation algorithms based on knowledge graph,as well as the problem of heterogeneous representation between data and knowledge graph in recommendation scenarios.In this thesis,we construct a collaborative knowledge graph and design a bilinear aggregation method,which makes full use of the rich semantic relationships between entity nodes in the graph and overcomes the problem of heterogeneous representation.Experimental results show that the algorithm can improve the capability of learning node feature representation.(2)Knowledge graph completion method based on graph attention network.To solve the problem of sparse data and incomplete entity information in knowledge graph in recommendation scenario,this thesis proposes a knowledge graph completion method based on graph attention network and conditional random field model,which can integrate neighborhood structure information.An encoder composed of relational attention network and conditional random field model is constructed to encode the interaction features between entities and relationships,and the neighborhood information is transmitted to the central node.Finally,the decoder is used to score and predict the triplet to complete the knowledge graph.Experimental results show that the proposed method performs well in knowledge graph completion tasks and can effectively improve the performance of recommendation under data sparsity.
Keywords/Search Tags:knowledge graph, recommendation system, graph neural network, attention mechanism, feature interaction
PDF Full Text Request
Related items