Font Size: a A A

Research On Link Prediction Method Of Knowledge Graph Based On Context Semantic Information

Posted on:2022-07-25Degree:MasterType:Thesis
Country:ChinaCandidate:X GaoFull Text:PDF
GTID:2518306329461164Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
Knowledge graphs can be seen everywhere in real life today,from virtual cyberspace to real language and encyclopedic knowledge systems.After several years of accumulation,in various fields of production and application,researchers and human experts have constructed a knowledge graph of various fields.After research,it is found that there is a large amount of missing information in the existing knowledge graph,and link prediction is an effective solution to make up for the missing information in the knowledge graph.Researchers also use many knowledge engineering techniques,such as Ontology,Framework and Rule-based knowledge graph link prediction methods.With the rapid development of deep learning technology,learning to achieve link prediction tasks based on knowledge graph embedding technology has become an important research paradigm in this field.Many models are proposed to represent the interaction between entities and relationships in the knowledge graph,that is,to capture richer entity semantic information.The existing models are not complete enough to model the long-term contextual semantics in the knowledge graph,and the longterm information between multiple triples expresses richer semantics,which is more important when completing link prediction tasks.Therefore,this paper proposes a knowledge graph link prediction method fused with contextual semantic information.This method uses a breadthfirst search sampling strategy to sample the entity relationship path composed of a sequence of triples,and input it to the long-term short-term memory network and feature extraction layer to capture the short-term and long-term semantic information in the triple path.Enhance the semantic representation capabilities of entity embedding and relationship embedding in the knowledge graph.And compared with many benchmark models on public real data sets,the knowledge graph link prediction method fused with contextual semantics proposed in this paper has achieved good performance indicators.It also verifies the effectiveness of the recurrent neural network architecture in the link prediction of the knowledge graph and the ability to capture the semantic information of the knowledge,which has superiority among many models.Through the analysis of existing models,it is concluded that the short-term and long-term semantic information in the knowledge graph is of great significance to the embedded representation of entities and relationships.Based on this design,the long-and short-term memory neural network captures the contextual information of the entity and enhances the embedding ability.Provide input data with rich semantic information for the knowledge graph link prediction task.The link prediction model of the knowledge graph is regarded as a prediction problem,and links are predicted in the context of part of the entity sample space.Design a specific search space for the proposed model to reduce the time and space complexity of the algorithm.When extracting the entity-relationship path,design the sampling strategy of the entity feature subgraph,so that the training data contains not only sufficient positive triple facts,but also high-quality negative triples to improve the generalization ability of the model.In the experimental stage,the experimental data set was analyzed and used for comparative experiments.The knowledge graph link prediction model that integrates contextual semantic information proposed in this paper has excellent performance on these data sets,which proves that the use of recurrent neural network structure to capture contextual semantic information is an effective and feasible implementation method.
Keywords/Search Tags:Knowledge Graph, Link Prediction, Representation Learning, Semantic Modeling
PDF Full Text Request
Related items