Research On Entity Linking Technology Based On DBpedia Knowledge Base

Entity linking,as a basic task in the field of natural language processing,maps a name string that appears in the text to a corresponding entity in the knowledge base.Entity linking is an important work in natural language processing technology,which provides a great help for effective semantic understanding.Entity linking technology has also been applied to many tasks of natural language processing,such as knowledge base update,knowledge base question and answer,information retrieval,machine translation and so on.Therefore,research on entity linking technology has an important significance.The authors propose an entity linking method based on DBpedia.The proposed method measures the association between entities of DBpedia knowledge base and constructs entity association graphs in different ways,builds a referential entity mapping dictionary to obtain referential candidate entities,computes the semantic similarity between entities and documents as a global feature,finally combines local features to perform entity linking iteratively.The main research contents of this thesis are as follows:(1).Four different methods for constructing entity association graphs are proposed,including the use of entity co-occurrence,Katz association,entity embedding similarity,and word vector similarity to measure the association between DBpedia entities.Then,four different entity association graphs are constructed.(2).According to the Wikipedia semi-structured information,a mention-entity mapping dictionary is constructed.Based on the dictionary and the mapping between Wikipedia and the DBpedia knowledge base,a set of candidate entities corresponding to the mention is found.Then,local features,containing literal similarity,context similarity and entity popularity are used to filter the set of candidate entities effectively.(3).Based on the entity association graph,a personalized EntityRank method is proposed.A personalized EntityRank vector is obtained for each candidate entity and document.The semantic similarity of the candidate entity is measured from the global perspective by comparing the personalized EntityRank vectors of the candidate entity and the document.In particular,the author use anchor entities to measure semantic similarity more accurately.Finally,the entity linking task is iteratively executed combining local features and global features.The effectiveness of the proposed method is verified experimentally on common datasets.Experimental results show that compared with the existing Rel-RW,AGDISTIS and DBpedia Spotlight entity linking methods,this method has improved accuracy,recall rate and F1 value.