Exploiting Hybrid Subword Information for Chinese Historical Named Entity Recognition

Chengxi Yan, Jun Wang

Abstract: Historical named entity recognition plays a very important role on the historical studies, especially for Chinese historical domain. This paper introduces a novel Deep learning (DL)-based approach with rich subword information of Chinese characters, which shows a powerful entity recognition of Chinese classical literatures. Our experiments on a manually constructed corpus of CMAG indicates that our model can achieve the best performance compared to other state-of-the-art approaches. The ablation studies on entity categories shows that the morphological information of Chinese characters can significantly improve the performance of NER models, especially combined with a gate-based neural unit GLU and the multi-head attention mechanism. This research provides some meaningful suggestions for automatic extraction of named entities in Chinese history, even extended to other low-resource domains.