Research On Knowledge Representation Learning Method Based On Data Expansion And Location Sensitivity

A very important part of artificial intelligence-knowledge graph,can organize,manage and understand the massive amount of information in the Internet in an orderly manner.The general knowledge graph structure is expressed in the form of triples(entity 1,relationship,entity 2).How to better represent the knowledge of the knowledge graph,so that the computing efficiency and operability are continuously improved,is the knowledge graph High-level requirements for overall quality.The knowledge graph representation learning method aims to represent the entities and relationships of the knowledge graph as low-dimensional,dense vectors,and is used for efficient semantic computing,and plays an important role in the construction,fusion and other aspects of the knowledge graph.The traditional knowledge graph representation learning model usually considers the existing facts in the knowledge graph,while ignoring the hidden semantic information in the knowledge graph.The current data-enhanced knowledge graph indicates that the learning model requires third-party tools or a large amount of manual intervention,and the reliability and stability of the data need to be strengthened.In addition,when the same entity appears in the head entity or tail entity position of different triples,the core information expressed is often different,and the existing model does not extract and use this part of the position-biased information.In view of the above problems,the main research contents of this article include:(1)A knowledge graph data expansion method based on reciprocal and symmetric relationship completion is proposed.This method can simply and efficiently expand the vast knowledge graph data set,directly increasing the effective triples,and greatly improving The effect of the existing general representation learning model;(2)A reciprocal relationship semantic logic constraint based on prior knowledge is proposed.This method strengthens the logical rationality of the entity pair representation of reciprocal relationship triples,thereby improving the effect of representation learning,and is based on the general representation learning model Show good universality;(3)A knowledge graph representation learning model based on position information is proposed,which effectively extracts the position preference information of entities in triples,adds this additional information to the representation learning process,and shows that in the link prediction experiment Good performance.Through standard link prediction experiments,it is shown that method(1)and method(2)have different degrees of enhancement and improvement to the previous representation learning methods,and the model proposed by method(3)has achieved good performance,surpassing most of the previous methods.The effect of the model.