EAGER: Enhancing Generative Event Extraction via Reinforcement Learning with Verifiable Rewards

2026-09-24

AI Summary

This study addresses the challenge of coordinating trigger identification, type classification, and argument extraction in end-to-end event extraction with large language models by proposing EAGER, a reinforcement learning framework. EAGER designs fine-grained, verifiable, multi-objective structured rewards and introduces Schema-Contrastive Advantage Estimation to effectively mitigate advantage collapse under sparse reward settings, thereby enabling efficient optimization of generative extraction policies. Experimental results demonstrate that EAGER significantly outperforms prompt engineering, supervised fine-tuning, and existing reinforcement learning methods across seven benchmark datasets.

Abstract

End-to-end event extraction remains challenging for large language models as it requires simultaneous identification of event triggers, classification of event types, and extraction of schema-grounded argument spans. We present EAGER, a reinforcement learning framework for generative event extraction that combines fine-grained verifiable rewards with Schema-Contrastive Advantage Estimation to alleviate advantage collapse under sparse binary rewards. Our reward design explicitly targets structural validity, extraction accuracy, groundedness, coverage, over-generation, and span precision. Experiments across seven benchmark datasets show that EAGER consistently outperforms prompting, supervised fine-tuning, and prior reinforcement learning baselines, achieving a substantial improvement over the strongest prior method. Results demonstrate that task-aligned verifiable rewards and contrastive advantage estimation substantially improve structured extraction.

Problem

• Event Extraction
• Large Language Models
• Reinforcement Learning
• Advantage Collapse
• End-to-End

Innovation

• Reinforcement Learning
• Event Extraction
• Verifiable Rewards
• Schema-Contrastive Advantage Estimation
• Large Language Models