Multi-Feature Fusion Transformer for Chinese Named Entity Recognition

Abstract

Recently, an approach to enhancing semantic and entity boundaries through lexicons has been used in Chinese named entity recognition tasks to improve recognition rate. However, this approach ignores the radicals of Chinese characters. As a kind of characters that evolve from pictographs, a lot of information is embedded in the radicals of Chinese characters. This paper proposes a novel Chinese named entity recognition method, named Multi-feature Fusion Transformer (MFT), which exploits the information of the radicals of Chinese characters. Our model is able to further enhance the semantic information by adding the radical information of Chinese characters while using word features. We have also improved the basic structure of the Transformer model to make the model perform better in entity recognition. Our model is compared with other mainstream models on Resume dataset and Weibo dataset, the experimental results show that our model is significantly better than other models, with F1 score of 95.77% and 64.38% for the respective dataset.