Semi-Supervised Named Entity Recognition Based On Deep Learning

Named entity recognition is a basic task in the field of natural language processing,and it is also the underlying technology in related fields such as machine translation,intelligent question and answer,knowledge graph and other applications.Deep learning technology has made great progress in named entity recognition due to its automatic and efficient feature expression and classification ability.Most of the existing deep learning named entity recognition methods are supervised training methods,which do not use unlabeled data to enhance the generalization ability of the model.Therefore,the semi-supervised named entity recognition method is studied in this paper.The main work of this article is as follows:Firstly,the research status of Chinese and foreign named entity recognition methods are investigated in this paper.The principle of named entity recognition based on statistics and deep learning is analyzed and summarized.On the basis of summarizing the advantages and disadvantages of each method,a semi-supervised named entity recognition method is proposed.A large number of unlabeled data are used to create similar sentences with labeled data to constrain the labeling sequence of sentences,which reduces the dependence of the model on labeled data.Experiments show that the constructed similar sentences can effectively correct the tagging sequence of sentences.The F1 value of the model in SIGHAN Bakeoff MSRA Chinese named entity recognition data set is 92.13%,which is 0.41% higher than that of the baseline model.Secondly,in order to use the long and short time memory network(LSTM)to extract the feature information far away from the sentence more accurately,a named entity recognition algorithm combining the Fixed-size Oradinally Forgetting Encoding(FOFE)and LSTM is proposed.By combining the static encoding method of FOFE with the dynamic forgetting method of LSTM,the ability of extracting sentence features of the model can be enhanced.The proposed model structure is used in the data sets of English and Chinese,with F1 values of 91.30% and 91.65% respectively,which validates the generality and effectiveness of the proposed method.In summary,this paper discusses and improves the named entity recognition method in semi-supervised and supervised way by using deep learning method,and further improves the model effect by adding external data and modifying the internal structure of the model respectively.The experimental results verify the effectiveness of the proposed method.