Research And Application Of Distributed Representation Of Knowledge Graphs Incorporating Comprehensive-Information

The distributed representation of knowledge graph is inspired by deep learning,which embeds knowledge into a low-dimensional dense vector space to reduce graph sparsity and improve computational efficiency.And there are relatively few works based on fusing auxiliary information to optimize knowledge graph embedding and have a large development space.Inspired by the theory of comprehensive-information and sememes,this paper innovatively proposes the Comprehensive-Information Knowledge Representation Learning Model Based on Sememe Feature(CIKRL-S),which constructs entity auxiliary information with grammatical and pragmatic information composed of sememe features to reduce the influence of data noise.In this paper,a comprehensiveinformation encoding mechanism based on a pre-trained BERT model is proposed in the entity comprehensive-information encoding stage,apply the encoding measure to the CIKRL-S model,and demonstrate its effectiveness in entity link prediction and triplets classification tasks.Many graph data do not contain the feature of sememes,and the sememes annotation is time-consuming and labor-intensive.To solve the above problems,we propose a Sememes-based Framework for Knowledge Graph Embedding with Comprehensive-Information(SCIKE)in combination with the predicted lexical sememes technology,comparing with previous research works,the model achieves optimal results in entity link prediction,triplets classification,and entity classification tasks.Finally,we combine various open-source technologies to build a prototype knowledge vector factory system,which aims to improve the efficiency of research on knowledge graph embedding.