Synonymous Entity Recognition Based On Entity Attribute And Content

Synonymous entity recognition is the task of finding different referral names of the same named entity in the dataset.Synonymous entity recognition can effectively solve data conflicts,remove redundant data,and promote the development of application fields such as question answering system.The traditional methods can only use limited textual information so that the recognition effect is not good.Compared with the traditional methods,the recognition effect of the search engine-based synonymous entity recognition method is greatly improved,but there are still some problems.Based on these problems,this dissertation carries out researches on the synonymous entity recognition task.(1)To solve the problem that the existing search engine-based synonymous entity recognition method does not make full use of entity information,a new similarity calculation method between named entities which is called as the VarSim function is proposed.The similarity method uses the entity page summary information returned by the search engine,and analyzes the hidden effective information in the content of the text information.Then combined with the feature fusion technology,the multiplicative feature fusion synonymous entity recognition method SER-mult-FF is proposed.SER-mult-FF method preserves and synthesizes the important authentication information of different named entity features to synonymous named entities and improves the recognition effect.Finally,experimental results verify the superiority of the SER-multi-FF method in synonymous entity recognition.(2)The existing search engine-based synonymous entity recognition method requires manual feature design and specific task expert knowledge,leading to great limitations of it.Based on the limitation,the Content-Aware Attributed Entity Embedding for Synonymous Named Entity Recognition method is proposed.This method combines Network Embedding and named entity information to construct two heterogeneous networks,and learns the form of the low-dimensional vector of named entity through the joint learning of the two heterogeneous networks.Then the distance of the named entities in the low-dimensional vector space is taken as the similarity measure between the named entities.The method can automatically extract valid semantic features from the original information of named entity attributes and text description content,without the need for manual feature design and domain expert knowledge,which improves the accuracy and efficiency of synonymous entity recognition.