Knowledge Graph Oriented Entity Alignment And Knowledge Completion

With the rapid development of artificial intelligence technology,Knowledge Graph,as an important basic technology in this field,is also gradually increasing in importance.Major enterprises and research institutions have also launched their own Knowledge Graphs.However,using only a single knowledge Corpus will lead to low coverage of information and information loss.Therefore,it is necessary to fuse multiple knowledge Corpus to construct a unified top-level knowledge base,which can effectively solve the problem of information missing.The effective entity alignment technology is the key to data fusion.On the other hand,Knowledge Graphs generally represent entities as nodes in a network and relationships as edges between nodes.We can infer potential knowledge from existing knowledge,and then complete the Knowledge Graph.Most of the existing knowledge Graphs complements use the translation model as the representative knowledge representation learning model,but the translation model does not make full use of the entity description text outside the knowledge base's own entity triple structure.The descriptive text of these entities can also provide a large amount of information for the knowledge Graphs completion.This article has carried on the thorough research from above two aspects,the main research content is as follows:(1)To solve the problem that the semantic information of unstructured summary text is ignored in current Chinese encyclopedia entity alignment methods,an improved WMD algorithm is introduced to propose a Chinese encyclopedia entity alignment method based on entity attributes and summary text information.In this dissertation,firstly,the attribute name and attribute value of entity attributes are normalized,and then the attribute similarity is calculated by editing distance algorithm.Secondly,the semantic similarity of abstract text is calculated by improved WMD algorithm.After synthesizing the two similarities,the entity alignment task is completed.Finally,experiments show that this method can improve the effect of entity alignment.(2)Since the existing representation learning model fails to make full use of entity description text,this dissertation establishes a knowledge representation learning model integrating text information.The text of the entity is encoded by using the depth Convolutional Neural Network to obtain the semantic information of the text representation.At the same time,the Trans H model is used to model the head-to-tail entities,and then a joint learning model is established based on the two.Experiments show that the model has better performance than the original knowledge representation model.(3)Constructing a Knowledge Graph of Qiandao Lake tourism.After crawling the relevant tourism information of Qiandao Lake scenic spots,hotels and restaurants that store websites such as Ctrip,ELong and Dianping,the multi-source data information is fused and aligned through the entity alignment method designed above.Secondly,the knowledge representation learning model integrating text information is used to complete the integrated knowledge base,further enriching the tourism Knowledge Graph of Qiandao Lake.