The Research On Entity Linking Based On Knowledge Graph

Entity linking refers to the process of mapping mentions in the text to an entity in the knowledge base.It is one of the key procedures in the field of knowledge graph and knowledge fusion.To address the issues of large storage space and long running time of graph-based entity linking algorithms in the long text field,two knowledge graph-based entity linking methods are proposed,Katz correlation-based entity linking method and confidence-based entity linking method included.The major contributions of this article are as follows:(1)An approach to candidate entity generation for batch threshold.In this article,the mentions are pre-classified.Then several adjacent ambiguous entities in the text are generated candidate entities through a method of a small number of times to reduce the scale of constructed graph,which further sovles the timeconsuming problem of graph-based entity linking methods in the long text field.(2)An entity linking algorithm based on Katz correlation.This paper uses the Katz association theory to compute the relationship of any two candidate entities,the problem of finding the closest subgraph in the candidate entity graph is converted to the problem of searching the largest submatrix of the relationship matrix between candidate entities.The hill-climbing algorithm is then applied to obtain the largest submatrix.The experimental results verify that this algorithm can significantly speed up the graph-based entity linking process in the long text.(3)An entity linking algorithm based on Confidence.This article calculates the confidence for each candidate entity,and since unambiguous mentions affect the selection of candidate entities for the entity to be disambiguated,the restart random walk is only conducted for the unambiguous mentions with a confidence of 1 to obtain the affinity between any unambiguous entities and other candidate entities.Furthermore,the affinity and confidence are combined for joint disambiguation.The experimental results prove that this algorithm can effectively improve the precision while ensuring the running speed at the same time.