Research On The Identification Of Nested Named Entities Based On Machine Reading Comprehension

With the advent of the information age in recent years,Internet social media has experienced rapid development.During this process,the information on the network has experienced an exponentially rapid growth.Text information is a very important part of it.Named entity recognition is the construction of dialogue robots.Basic tasks for advanced natural language processing applications,such as automatic summarization.The research of naming entity recognition has promoted the practicalization of natural language processing technology.Most of the research focuses on non-nested named entities,while ignoring nested named entities,resulting in a large amount of information loss,especially in the text with more names,institutional names,and biomedical words containing a large number of complex nested named entities.The identification of nested named entities can obtain a large number of more layered structured information,improving the quality and quantity of information extraction.This paper focuses on the deep learning method of the identification of nested named entities,the main content of the work and the results obtained at the stage are as follows:(1)A baseline system for the identification of named entities based on BiLSTM-CRF’s integration of entity knowledge is constructed,and the role of model integration into entity knowledge is studied,and the improvements needed for the identification of nested named entities are analyzed.(2)On the basis of the attention machine reading comprehension model,an entity query nested named entity recognition algorithm model based on the attention mechanism is constructed,and the entity representation method of the pointer position is designed.which performed.Analyze the main components of the model through ablation experiments.(3)Analyze the application of the BERT pre-training model in machine reading comprehension tasks,construct a named entity recognition model based on pre-trained machine reading comprehension,optimize the prediction layer,and do not need to adapt to different data sets.Entity recognition is unified with nested named entities,and achieves better performance on public data sets.