Research On Nested Named Entity Recognition Integrating Syntactic Information

Nested named entity recognition is a key and difficult task in natural language processing.The improvement of the performance of this task can drive the improvement of the performance of a series of downstream tasks,which has high application value and social value.With the development of deep learning,many nested named entity recognition methods based on deep learning have emerged in recent years,which further improves the performance of nested named entity recognition.However,the existing methods ignore the relationship between syntactic information and nested named entity recognition,and pay insufficient attention to syntactic information.In addition,due to the particularity of nested named entity recognition task,it is often necessary to design complex models to solve it,resulting in the decline of the interpretability of the model.Therefore,this paper studies the nested named entity recognition method integrating syntactic information,and makes full use of syntactic information to improve the performance of nested entity recognition and the interpretability of the model.The main contents of the study include the following:(1)Aiming at the high coincidence of sentence phrase structure and nested entity structure,a nested named entity recognition method based on phrase expansion is proposed.In this method,the dependent syntactic information is fused into the vector representation of words through graph neural network,and then the phrase syntactic information is modeled by tree LSTM.All phrase structures are extended to candidate nested named entities after boundary regression,and finally classified.Experiments show that the integration of syntactic information is helpful for nested named entity recognition.After integrating phrase syntactic information through phrase expansion,the performance of the model is effectively improved.(2)Aiming at the problem of syntactic information errors in neural network modeling,a nested named entity recognition method based on attention mechanism integrating syntactic information is proposed.This method formulates relevant rules to accurately extract syntactic information,and integrates phrase syntax and dependency syntax into the model through attention mechanism to obtain important syntactic information,strengthen the connection with nested entity recognition and enhance the interpretability of the model.Experiments show that on ace2005 and genia corpora,the nested named entity recognition method based on attention mechanism integrating syntactic information achieves better recognition effect than the existing methods,and further improves the performance of nested named entity recognition.