Research On Representation Learning For Knowledge Graph

With the development of science and technology,we gradually entered an era of information and intelligent.The amount of data that is being created and stored on a global level is almost inconceivable,and it just keeps growing.Mobile internet has become the most effective and convenient information acquisition platform.How to obtain effective information from massive data has become a major problem in many fields.Therefore,representation learning for knowledge graph has become a major topic in the field of artificial intelligence.Knowledge graph is a graph structure of "entity" and complex semantic of "relation" between these two entities in the real world.Representation learning aims to project both entities and relations into a continuous low-dimensional vector space,and then to expand knowledge analysis to numerical computation.In recent years,researchers have proposed a variety of representation learning models represented by TransE to learn the correlation between entities and relations in a knowledge graph,and have achieved remarkable results.However,there are still many limitations,for instance,TransE is not good at dealing with complex relations and the fusion of multi-source information.To solve these two issues,the main research content of this paper is as follows:(1)To address the limitations of TrasnE for modeling complex relations,a representation learning method with multi-translation principle is proposed.Firstly,the model parameters are constantly adjusted during the training process.Secondly,according to different relation categories,the principles of translation are different,which effectively solves the limitations of modeling complex relations.(2)The existing representation learning models usually use the same step to train the entity and relationship with different complexity,which can't accurately distinguish the relation of different complexity.In order to balance the influence of step size on the relation and entity of different complexity,we define a dynamic step size based on the complexity of entities and relations to solve the effect of heterogeneity and imbalance of knowledge graph.(3)In view of the problem that the current model fails to make full use of knowledge-related information,a representation learning method with the fusion of text description information about entities and relations is proposed.A novel tagging scheme is used to jointly extract entities and relations,and a balance factor is defined to combine text description information with structured information,which can better represent the data in a knowledge graph.