EventRL：通过为大语言模型引入结果监督来增强事件抽取
EventRL: Enhancing Event Extraction with Outcome Supervision for Large Language Models
AI总结：
提出强化学习方法 EventRL，通过结果监督和特定奖励函数优化大语言模型的事件抽取，有效缓解幻觉与指令遵循问题，在多模型上显著优于 FSP 和 SFT 等基线方法。
摘要：
在本研究中，我们提出了 EventRL，这是一种为增强大语言模型（LLMs）事件抽取能力而开发的强化学习方法。EventRL 利用带有特定奖励函数的结果监督，以应对 LLMs 中普遍存在的指令遵循和幻觉等挑战，这些挑战表现为事件结构的不匹配以及生成未定义的事件类型。我们在包括 GPT-4、LLaMa 和 CodeLLaMa 模型在内的多种 LLMs 上，将 EventRL 与现有方法（如基于 GPT-4 的 Few-Shot Prompting (FSP) 和 Supervised Fine-Tuning (SFT)）进行了对比评估。研究结果表明，EventRL 通过提升事件识别和结构化的性能，显著优于这些传统方法，特别是在处理新颖事件类型方面。本研究强调了选择奖励函数的关键作用，并展示了融入代码数据以优化事件抽取的益处。虽然增加模型规模会带来更高的准确率，但保持泛化能力对于避免过拟合至关重要。
英文摘要：
In this study, we present EventRL, a reinforcement learning approach developed to enhance event extraction for large language models (LLMs). EventRL utilizes outcome supervision with specific reward functions to tackle prevalent challenges in LLMs, such as instruction following and hallucination, manifested as the mismatch of event structure and the generation of undefined event types. We evaluate EventRL against existing methods like Few-Shot Prompting (FSP) (based on GPT4) and Supervised Fine-Tuning (SFT) across various LLMs, including GPT-4, LLaMa, and CodeLLaMa models. Our findings show that EventRL significantly outperforms these conventional approaches by improving the performance in identifying and structuring events, particularly in handling novel event types. The study emphasizes the critical role of reward function selection and demonstrates the benefits of incorporating code data for better event extraction. While increasing model size leads to higher accuracy, maintaining the ability to generalize is essential to avoid overfitting.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
