Chinese Event Extraction for Epidemic Prevention and Control Domain

This paper focuses on the task of event extraction in the context of epidemic prevention and control. However, this research area still faces numerous challenges. One significant issue is the absence of a dedicated dataset for events related to epidemic prevention and control. Additionally, the existence of long triggers and multiple triggers complicates the process, leading to inaccuracies and omissions in machine extraction. Furthermore, the uneven distribution of event arguments has an impact on the accuracy of extraction outcomes. To address these issues, this paper first develops a dataset for key events in epidemic prevention and control, named EEPCD. It then introduces A-DPETE, an algorithm that uses dependency syntactic analysis to extract event triggers. Finally, the paper introduces the EM-TFEEA model, an event argument extraction model that integrates trigger features. Employing the grouping extraction principle, this model significantly enhances the accuracy of event argument extraction. Experimental results show that the proposed methods, applied to both the EEPCD dataset and the ACE2005 Chinese dataset, outperform conventional techniques. The accuracy, recall, and F1 scores for event trigger extraction maximally improved by up to 6.0%, and for event argument extraction, these metrics maximally rose by up to 3.0%.