Chinese Named Entity Recognition Based on Multi-feature Fusion

Pre-trained language models usher in a new era of named entity recognition, but more additional relevant knowledge is needed to improve its performance on specific problems. In particular, in Chinese government named entity recognition, most entities are lengthy and have vague boundaries, and this entity length and boundary uncertainty makes the entity recognition task difficult or incorrectly identified. To address this problem, this paper proposes a Chinese named entity recognition model based on multi-feature fusion, in which lexical features, word boundary features and pinyin features are fused together through a multi-headed attention mechanism to enhance the model’s semantic representation of government texts. Meanwhile, this paper also studied the contribution of different features to entity recognition, and finds that pinyin features have unique advantages in recognising government entities. This study provides new ideas and methods for the research and application of Chinese governmental entity recognition, and also provides some insights into the problem of named entity recognition in other language domains. The experimental results show that the model proposed in this paper has better performance compared to the baseline model.