Research And Implementation Of Graph Based Chinese Collective Entity Linking Algorithm

The definition of entity linking is to link named mentions in the plain text to their referent entity in the knowledge base,and return NIL mark if the corresponding entity is not in the knowledge base.In the wave of rapid development of Internet technology,the content of knowledge stored in the network is growing at an exponential rate,such as micro-blog,forums,news and so on.How to extract useful information from these massive data put forward more stringent requirements for natural language processing technology.Among them,named entity is an important part of natural language text,and how to quickly and accurately identify the specific object refers to the named entity in natural language text(i.e.entity linking),is of great help to understanding the meaning of the text.Therefore,entity linking has a profound research significance and extensive application value for the development of natural language processing.Based on the existing research of entity linking at home and abroad.In this thesis,we presents two kinds of graph-based collective entity linking algorithms for Chinese,including Context Graph-based Collective Entity Linking Algorithm(CGCEL)and Consistent Collective Entity Linking Algorithm(CCEL).In summary,the major contributions of this thesis are described as below:1.This thesis presents a new method of referent graph construction,this method divides the semantic association between entities into direct semantic relations and indirect semantic relations,which can not only guarantee the strength of semantic relation between entities,but also guarantee the integrity of referent graph.2.In order to further improve and revise the referent graph,an incremental evidence mining algorithm is proposed in the stage of referent graph construction.This algorithm can reduce the dependency of entity linking algorithm on local knowledge base and improve the linking accuracy by using the background knowledge of the entity provided by the external knowledge base.3.This thesis presents a context graph-based collective entity linking algorithm,called CGCEL.The algorithm model the semantic relevance between candidate entities in the form of the graph and then find out the importance of the candidate entity to get the correct linking object.The experimental result shows that the algorithm can solve the problem of entity linking elegantly.4.This thesis presents a consistent collective entity linking algorithm,called CCEL.Through a more in-depth consideration of the semantic correlation between candidate entities and comprehensive consideration the relevance between candidate with entity mention and the semantic correlation in candidate entities with documents,the algorithm can minimize the noise generated by the wrong candidate entities and improve the distinction of similar entities,so as to improve the effect of entity linking algorithm.