Research on the Semantic and Structure Fusion-Based Knowledge Graph Completion Model
[Objective]This paper proposes a knowledge graph completion model with semantic and structural information.It improves the completion,reliability,and quality of the knowledge graph.[Methods]First,we used a pre-trained language model to enhance the knowledge graph's embedded text and context data.Then,we captured the semantic information of entities and relationships.Third,we constructed an entity-relationship matrix to map the network structure of the knowledge graph and obtain each entity's neighborhood information and relationship constraints.Finally,we integrated the potential data to train the model and predict the missing entity of the knowledge graph.[Results]Compared to the baseline method,the proposed model's Hits@3 metric improved by 0.5％,0.6％and 0.6％on the FB15k-237,WN18RR and UMLS data sets,respectively.[Limitations]Due to the language models'information representation ability limits,we cannot further improve the knowledge graph by completing tasks with the help of multimodal data.[Conclusions]The proposed method can perform better for the knowledge graph completion task,promoting the knowledge graph's development and its downstream application.
Knowledge Graph CompletionPre-Training Language ModelNatural Language ProcessingDeep Learning
NETL
NSTL
万方数据
