Multitask Learning for Chinese Named Entity Recognition

Named Entity Recognition (NER) for Chinese corpus such as social media text and medical records is a grand chanllenge as the entity boundary is not easy to be accurately clarified. In this work, we describe and evaluate a character-level tagger for Chinese NER, which incorporates multitask learning, self-attention and multi-step training methods to exploit richer features and further improve the model performance. The proposed model has achieved 90.52% strict F1 on the Electronic medical records dataset (CCKS-NER 2017), which is the best single model at present. In addition, we also conducted experiments on a Chinese Social Media dataset and the CCKS-NER 2018 dataset, whose results illustrate the effectiveness of the proposed method for Chinese Named Entity Recognition task.