Research On Tibetan Named Entity Recognition Based On Deep Learning

Natural language processing studies the language problems in the communication between humans and computers,and it is the key to the information construction of various languages.The research on Tibetan natural language processing is a symbol of the development of Tibetan information technology.It is of great significance to carry forward Tibetan culture,expand Tibetan cultural influence,promote ethnic cultural exchanges,enhance the overall level of Tibetan informatization,and promote economic development and social progress.Tibetan named entity recognition is one of the key technologies in Tibetan natural language processing,and its recognition performance will directly affect the performance of subsequent processing tasks or related applications.At present,the research of Tibetan named entity recognition is mainly based on traditional machine learning methods.Those methods require construct feature engineering according to linguistic knowledge.Such processes are cumbersome and the system portability is poor.The deep learning methods can avoid these problems by self-learning features,and can achieve performance equivalent to or better than current traditional machine learning methods.Therefore,this article carried out relevant research on Tibetan named entity recognition based on deep learning.The main research content is as follows.1.Perform data cleaning and data labeling on the Tibetan part-of-speech corpus,and make a Tibetan entities corpus for named entity recognition research.The final corpus contains 58348 sentences.2.Based on RNN and CNN model,we construct Bi-LSTM-CRF,IDCNN-CRF and improved IDCNN-CRF model to train character vectors with using word vector technology.By comparing the experimental results of the traditional machine learning model and the deep learning model on the same Tibetan corpus,the effectiveness of the method based on deep learning on Tibetan named entity recognition task is verified.Among them,improved IDCNN-CRF model has a better entity recognition effect with the F1 value 80.16%.3.In order to learn the semantic relationship between the named entities and the tags completely,we further improved the Tibetan named entity recognition model by integrating attention mechanism with the established model framework.Using attention mechanism is mainly to strengthen the semantic relationship between the named entities and the tags,and to improve the effect of named entity recognition.The experimental results show that the model incorporating attention mechanism can further improve the recognition effect of the named entities.