Research on Chinese Event Extraction Method Based on RoBERTa-WWM-CRF
Zhenghan Li, Nanchang Cheng, Wenchao Song · 2021
To extract the event information contained in the Chinese text effectively, this paper takes Chinese event extraction as a sequential labeling task, and proposes a method to extract events based on the combination of RoBERTa-WWM (A Robustly Optimized BERT Pre-training Approach-Whole Word Masking) and Conditional Random Fields (CRF). This method uses RoBERTa-WWM to generate semantic representation with prior knowledge, and then inputs them into the Conditional Random Fields (CRF) model. The argument is predicted by the output label sequence. The experimental results show that this method can effectively improve accuracy, recall, and F1-score on the Chinese event extraction dataset DUEE1.0, which Baidu recently released, and improve the performance of event extraction in Chinese text.