Chinese Medical Named Entity Recognition using CRF-MT-Adapt and NER-MRC

Medical Named Entity Recognition is a fundamental component of understanding the medical free-text notes in Electronic Health Records, and it has become a popular research topic in both academia and industry. The China Conference on Knowledge Graph and Semantic Computing (CCKS) organizes a challenge for Medical Named Entity Recognition, aiming at extracting medical entity mentions and categorizing them into pre-defined classes. We propose a Multi-Task sequence labeling model with Adaptive Loss Weighting (CRF-MT-Adapt) to address the issue of low recall and a Named Entity Recognition model based on Machine Reading Comprehension (NER-MRC) to address the issue of long-span entity mentions. We experimentally demonstrate the state-of-the-art performance of the two proposed models and the ensemble even surpasses the strong baselines by at least 2% F-score. On the official test set, our best submission achieves an F-score of 90.51% and 95.96% under strict and relaxed criteria respectively.