Fine-Grained Chinese Named Entity Recognition Based on MacBERT-Attn-BiLSTM-CRF Model

Named Entity Recognition (NER) is to identify pre-defined types of entities (such as people, organizations, locations) from texts. NER is a key component of information retrieval, relationship extraction and other tasks in practice. However, traditional entity recognition models have a series of problems: high cost of artificial feature design, weak model robustness, and coarse granularity of entities. To solve these problems, we propose a model called MacBERT-Attn-BiLSTM-CRF based on pre-trained language models. The experiments on a fine-grained Chinese NER dataset show that our model outperforms existing models significantly.