Research On Knowledge Representation Learning Fusing Of External Text Information

The knowledge graph is an important part of artificial intelligence.It can better organize,manage,and understand the vast amount of information on the Internet and present it to people in a structured form with its powerful semantic processing capabilities and open interconnection capabilities.The triple "entity,relationship,entity" is a general representation of knowledge graph.Entities are connected to each other to form a semantic network and stored in the knowledge graph in symbolic form or network structure,but such representations exist.The problems of sparse data and poor computing efficiency,so how to better represent the knowledge in the knowledge graph is the key to improving the quality of the knowledge graph.With the popularity of deep learning,representation learning technology has gradually attracted people's attention.Knowledge graph-oriented representation learning has become the basis for constructing high-quality knowledge graphs.Knowledge representation learning aims at knowledge graph-oriented representation learning,mapping entities and relations in the knowledge graph to a low-dimensional dense vector space,under which the entities and relations are represented in vector form,which is convenient for calculating And discover deeper semantic connections.This technology can significantly improve computing efficiency,alleviate the problem of sparse data,and achieve the fusion of heterogeneous information from multiple sources.It has greatly improved the quality of downstream tasks such as knowledge base completion and knowledge reasoning.Previous knowledge representation learning techniques only used the structural information of the triples themselves,but the knowledge graph also contains a large amount of entity and relationship description information,category information,and even a large amount of text on the Internet that has not been added to the knowledge graph Information,which can improve the discriminating ability of knowledge representation,has not been discovered and used.Based on the above reasons,thisarticle will use the entity description information of the knowledge graph to enhance the representation of the entity,thereby improving the quality of the knowledge representation.The main work is as follows:(1)In the learning process of entity representation,the text description information of the entity is introduced.Entity description information is a brief introduction to the entity.Through the description information,the attributes and definitions of the entity can be known.The textual representation of the entity description information is combined with the structural information of the entity to make the representation of the entity contain more semantic information.(2)A text supervised representation learning model TBTS is proposed.For the first time,the Transformer structure is introduced into knowledge representation learning.Part of the Transformer structure is used to learn the description of the description text of the entity.Maximize the use of information in different subspaces.(3)Use the idea of adversarial generative networks,and use other models as generators to provide models with better negative samples,solve the "false negative example" problem and "zero loss problem",thereby improving the efficiency of model training.The model performs entity link prediction and triad classification task experiments on two data subsets,FB15 K and WN18.The experimental results show that the two evaluation indicators are superior to other comparison models,proving that the model can use text information to perform existing representation Promotion.