Research On Chinese Named Entity Recognition Based On Graph Network

Natural Language Processing(NLP)is an important part of the field of artificial intelligence.This direction includes applications such as text processing,information extraction,sentiment analysis,information retrieval,machine translation,and question answering systems.The basic work of these applications is named entity recognition(NER),and the effect of subsequent work will be directly affected by the recognition results.The task of Chinese named entity recognition is to identify the entities in the text to be processed that represent specific actual meanings.At present,the commonly used solution is based on deep learning,but the traditional deep learning model has a large amount of parameters and cannot effectively capture long-distance features.The parallelism of GPU cannot be fully utilized in the training process,and traditional language representation models cannot solve Chinese.The problem of ambiguity in the word,caused difficulties for the later work.In response to these problems,this article proposes the following two solutions:(1)A Chinese named entity recognition method based on the BSTTC(BERT-Star-Transformer-Text CNN-CRF)model is proposed.In order to avoid further propagation of errors caused by word segmentation,the concept of single-character vectors is used in the language representation module,and the BERT model is used to solve the problem of ambiguity in Chinese.In the feature extraction stage,in order to solve the problem of high computational memory overhead and low training efficiency of the classic Transformer model,a lightweight Star-Transformer model is used as the basic model in the feature extraction module to stack to obtain a representation containing internal semantic information and sentence local features Matrix,and added Text CNN as a global feature extractor to make up for the lack of long-distance feature extraction of the Star-Transformer.Finally,the feature vector sequence output by the deep learning model is used as the input of the conditional random field to obtain the best label discrimination result of the sentence sequence.By designing multiple sets of comparative experiments,the effectiveness of each part of the method is proved,and the results of comparison with existing models also show that this method can effectively improve the effect of Chinese named entity recognition.(2)A Chinese named entity recognition method based on two-way graph attention network is proposed.In order to solve the problem that the traditional feature extraction model Bi LSTM cannot achieve parallel operation and the CNN model is difficult to extract global information,the feature extraction process of Bi LSTM and CNN is simulated by running the graph attention mechanism in parallel in the two graph structures,which can not only capture the long-distance feature of the sentence,and make full use of the parallelism of the GPU.In addition,considering that the sequence information of the sentence has a certain degree of influence on the final label discrimination result,a two-way mechanism is used in the graph attention network.Experimental results show that compared with traditional methods,this method has improved recognition accuracy and F1 value.