Conference Paper Published

CCL25-Eval任务四系统报告：基于RAG与谓词相似性方法的叙实性检测智能体

(System Report for CCL25-Eval Task 4: A Factivity Detection Agent Based on RAG and Predicate Similarity Methods)

Wang, Y., Yang, Q., Liang, K., Yang, Y., Zhai, Y., & Huang, C.-R. (2025). CCL25-Eval任务四系统报告：基于RAG与谓词相似性方法的叙实性检测智能体. In Proceedings of the 24th China National Conference on Computational Linguistics (CCL 2025), 152-156.

URL: https://aclanthology.org/2025.ccl-2.18/

摘要 Abstract

本文聚焦于“叙实性推理”任务，即判断语言中事件真实性的语义理解能力。该任务不依赖外部知识，而基于语言结构本身进行推理，对当前大语言模型(LLMs)提出挑战。为解决模型在叙实性漂移、多义词处理等方面的不足，作者提出一种结合RAG(检索增强生成)与谓词相似性的方法，构建了一个融合参数化与非参数化知识的叙实性检测智能体系统。该系统通过分步提示与知识库支持，实现了更高的一致性、准确性与可解释性，在评测任务中取得了0.9240的稳健表现。

This paper addresses factivity inference—determining event truth based on linguistic cues. To improve large language models’ performance on this task, the authors propose an agent system combining Retrieval-Augmented Generation (RAG) and predicate similarity. The method integrates structured and contextual knowledge, enabling accurate, consistent, and interpretable truth-value judgments. It achieves a strong evaluation score of 0.9240.

关键词 Keywords

叙实性 Factivity, RAG, 谓词相似性 Predicate similarity, Agent