Research On Chinese Named Entity Recognition Based On Deep Learning

The internet industry of our country has made great progress in recent years,and a large amount of Chinese text which contain a lot of valuable information has been produced in this process.Named entities are important semantic knowledge in texts,which affect the effects of many higherlevel tasks in the field of natural language processing,including machine translation,question answering systems,sentiment analysis and search engines.Therefore,identifying and classifying named entities in text accurately has become a key basic research task gradually.Chinese named entity recognition can extract entity information from Chinese text,which can lay a foundation for related research in the field of Chinese natural language processing.However,Chinese named entity recognition research is facing difficulties in dividing the boundaries of Chinese entities,word ambiguity,limited labeling data and so on.Traditional named entity recognition research methods are mainly based on rules or statistical models,both of them require manual participation in formulating rules or constructing features and rely on the linguistic knowledge of the field heavily.The methods based on deep learning do not require human intervention,and the deep learning model can extract sentence features automatically.In particular,the emergence of distributed representation based on neural networks has brought strong development momentum to the research of named entity recognition.Aiming at the current difficulties and challenges which Chinese named entity recognition is facing,this paper not only studies the related work of deep learning in this field deeply but also makes summary,exploration and innovation.The main research contents of this paper are as follows:1.Aiming at the problem that the Chinese named entity recognition method based on the character level cannot make full use of the sentence and word information,this paper designs a Chinese named entity recognition model that integrates word information.By introducing an external dictionary,the model incorporates the information of potential words in the sentence into the character-level representation,avoiding the problem of word segmentation error propagation.In addition,the Transformer network is used to encode and analyze the vector of characters and words and position relationship information,then the optimal tag sequence is calculated by CRF layer.Experiments were carried out on the Chinese named entity corpus CLUENER2020.The experimental results showed that the accuracy rate reached 82.46%,the recall rate reached 83.14%,and the F1 value reached 82.80%,indicating that the method of fusing word information can improve the recognition effect of Chinese named entities.2.Based on the ideas of generative adversarial network and multi-task learning,this paper studies the effect of adversarial multi-task learning on improving the performance of Chinese named entity recognition.Through adversarial training Chinese word segmentation task and Chinese named entity recognition task,the word boundary parameter information of Chinese word segmentation task is shared to named entity recognition task while avoiding the influence of task private feature noise on shared feature space.In the experimental part,this paper uses two Chinese word segmentation data sets,PKU and MSR.Compared with the benchmark model,the F1 value has been greatly improved.This verifies the effectiveness of the model in this paper and shows that the adversarial training can help improve Chinese Named entity recognition effect.In addition,it is also verified that the attention mechanism has a strong ability to capture long-distance dependencies in sentences.