Research On Chinese Named Entity Recognition Algorithm Based On Textual Information Perceptual Fusion

With the rapid development of computer technology and social informatization,all kinds of information in life have been digitally stored and displayed,and extracting the required physical information from hundreds of millions of text resources has become a huge problem faced by people.Named Entity Recognition(NER)as an effective entity extraction method can solve the above problems.In the field of Chinese named entity recognition,named entity recognition methods based on deep neural networks have been widely studied.In this thesis,aiming at the problem of insufficient utilization of the overall information of the dataset by existing methods,a named entity recognition method based on Textual Information Perception Fusion(TIPF)is proposed.Firstly,aiming at the problem that the previous named entity recognition method based on deep neural network only focuses on the local feature information of the statement to be recognized,and ignores the influence of the global feature information of the dataset on the recognition effect,a feature fusion method of Textual Information Memory Perception(TIMP)is proposed.This is because different statements in the same dataset are usually from the same article,and there is a certain information association between different statements,so the feature extraction module of the named entity recognition method integrates the global feature information of the dataset,so that when the feature extraction of the recognition statement is performed,the overall information of the dataset(that is,other statement information)can be effectively considered to the recognition impact of the current statement,and the overall information of the dataset can be utilized.Secondly,in order to solve the problem of the fusion accuracy of global feature information in the process of named entity recognition,we introduce a selective kernel network based on the TIMP method,and propose an adaptive fusion method for text information,which generates adaptive weights of local feature information and global feature information according to the contribution of global feature information to the current character recognition results,and realizes the adaptive fusion of local feature information and global feature information according to this weight,so as to achieve the purpose of accurate fusion of the overall information of the dataset.Through the above two methods,the full extraction and accurate fusion of text context statement information are realized,and the information is applied to the named entity recognition process,which effectively improves the accuracy of named entity recognition.Finally,three real data sets show that the proposed method improves the evaluation indicators such as accuracy,recall and F1 value,and is better than other models.In terms of model training time,the proposed method can achieve higher recognition accuracy than other models by using shorter training time.