Entity Linking Based On Knowledge Base

Because of the ambiguity of natural language,it is a key problem to search the entity's semantic disambiguation,and the entity linking task based on the semantic disambiguation has emerged.The entity linking task is the process of extracting the named entity,including person name,geography name,organization name.At the same time,link the entity to the knowledge base.In this paper,the problem of entity link is studied.The main contents are as follows:1.An extension method based on word embedding is proposed.In this method,cbow model is used to train the word embedding,and then theextended word is used as the query,which can make up the deficiency of the method based on rule.Meanwhile,the method can improve the accuracy of the recall to the candidate entities.2.A ranking model based on word embedding is proposed to realize the entity linking.In the feature extraction phase,not only the LDA theme similarity between documents is added,but also the text similarity based on word embedding is added.In the context of the entity,the dimension of the vector is no longer the high frequency word,but the similarity word based on the word embedding.Finally,the learning to rank model is used to link the query to the corresponding candidate entities.3.A semantic web model based on word embedding is proposed to realize the entity linking.All entities in the document and the candidate entities are used as nodes to construct the semantic web,and the weights between nodes and nodes are represented by the semantic relations between words and words.In this paper,the experiment use the data sets of the entity linking task in 2014.The results show that the learning to rank model can make the F1 value to 0.71,and the semantic web model can make the F1 value to 0.739,which has good effect.This experiment shows that the two ideas of the entity linking method can better solve the problem of entity linking.