A Chinese Named Entity Recognition Method Based on Fusion of Character and Word Features

Named entity recognition is the upstream task in natural language processing tasks and is the basis for carrying out other downstream tasks. To enhance the effectiveness of the model for named entity recognition, a character vector with semantic features is obtained using BERT as the underlying encoder, followed by contextual features of the text sequence via BILISTM. In the Chinese named entity recognition task, both words and characters are equally important to the text, so a FLAT network is embedded to fuse word and character features. The network uses clever relative position encoding to preserve the location information of the input token, and generates potential word and character vectors to be added to the model for training. Experimental results show an increase in F1 values of 1.86% and 1.47% on the Resume and self-annotated news corpus datasets, respectively.