Design of question answer prompt for language model-based information extraction

Summarise with VIRAT
Authors: Yini Yang
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Information extraction is a critical yet resource-intensive task, necessitating vast amounts of high-quality annotated data. Addressing this challenge, my research proposes an event data generation method based on Large Language Model (LLM), which excels in low-resource event extraction scenarios by formulating the task as a guided conditional generation process. My model leverages end-to-end method to extract entities within passages into structured format adhering to predefined patterns with refined prompt engineering. Instead of manually crafted prompts, my model iteratively optimizes event-specific templates that incorporate event type definitions, candidate pools, and multi-level instructions. This innovation allows model to learn joint trigger-argument representations effectively, even with sparse training data. Empirical results demonstrate that my model performs well in low-resource event extraction, benefiting from its ability to utilize fewer annotated examples and capture intricate event dependencies. Moreover, human assessments corroborate the superiority of model-generated data, which often surpasses the performance of human-annotated data, offering a promising avenue for minimizing human labeling efforts in real-world applications.

Fields of Study & Subjects

Topic Modeling Natural Language Processing Techniques Multimodal Machine Learning Applications

Publication details

Journal / source
DR-NTU (Nanyang Technological University)
Publication date
1 January 2026
Language
EN
Document type
Thesis