Research On Nested Named Entity Recognition In Geographical Domain Based On Hierarchical Tagging

Named entity recognition(NER)is a basic task of constructing the geographical question answering system of college entrance examination.Researches on NER play an important role in promoting the practical use of natural language processing technology in recent years.However,most researches on NER only focus on non-nested named entities in the text,but ignore nested named entities,resulting in the loss of a large amount of information.In addition,Chinese corpus of nested named entities is extremely scarce,especially in Chinese geographical domain.And there is no nested named entity recognition(NNER)research in Chinese geographical domain.Based on the above problems,this thesis analyzes the characteristics of named entities in geographical domain and studies geographical NNER in order to fully identify the named entities in geographical text and to improve the performance of traditional NER.The main work is as follows:(1)With respect to the problem of lacking corpus in geographical domain,this thesis annotates Chinese high school geographical textbooks using hierarchical tagging method.This thesis constructs a geographical nested named entity corpus with 9758 named entities.It provides data for NNER research in Chinese geographical domain.(2)With respect to the problem that there is no NNER research in Chinese geographical domain currently,this thesis hierarchically models NNER task for the first time in geographical domain and studies the NNER task adopting Conditional Random Fields(CRF)model and Bidirectional Gated Recurrent Unit(Bi-GRU)neural network model,respectively.Experiments show that recognizing nested named entities using hierarchical tagging method greatly enhance the performance of geographical NER.The F1 score of the two kinds of models based on hierarchical tagging reach 60.13% and 50.28%,respectively,and they both achieve higher recall than the original baseline which does not use hierarchical tagging method.In addition,when using hierarchical tagging method in geographical NNER task,the overall performance of CRF model is better than that of Bi-GRU neural network model.However,the performance of Bi-GRU neural network model in recognizing some categories is better than that of CRF model.