基于多任务学习的中文命名实体边界感知神经网络模型

Named entity recognition for Chinese corpus has been a great challenge. In order to improve the result of Chinese named entity recognition, this paper propose a boundary-aware neural model for Chinese named entity recognition based on multitask learning. In this paper, the traditional named entity recognition task is decomposed into two sub-tasks: entity boundary detection and entity category classification. Through the decompositionof task, the pertinence of the model for the boundary recognition of Chinese named entities is improved. The model calculates the loss separately, and uses the combination of entity features and entity category comprehensive expression information to improve the accuracy of entity classification. Through experiments, it is found that the model results have improved on multiple data sets, and active SOTA on the data sets including Chinese MSRA, OntoNotes 4.0, Resume and Weibo, which are +0.33, +1.12, +0.76, and +2.83. The results verify the effectiveness of the model in this paper and provide new ideas for the task of Chinese named entity recognition.