The Art of Prompting: Event Detection based on Type Specific Prompts

• DOI: 10.18653/v1/2023.acl-short.111
• Publication Date: 2023-08-05T00:57:42Z
• Authors: Sijia Wang; Mo Yu; Lifu Huang
• Keywords: Complex Event Processing

We compare various forms of prompts to represent event types and develop a unified framework incorporate the type specific for supervised, few-shot, zero-shot detection. The experimental results demonstrate that well-defined comprehensive prompt can significantly improve detection performance, especially when annotated data is scarce (few-shot detection) or not available (zero-shot detection). By leveraging semantics types, our shows up 22.2% F-score gain over previous state-of-the-art baselines.