首页|期刊导航|华侨大学学报(自然科学版)|基于事件语义学的开放域无监督事件抽取

基于事件语义学的开放域无监督事件抽取OA

Unsupervised Open Domain Event Extraction Based on Event Semantics

中文摘要英文摘要

面向开放域、缺乏任务特定标注的数据场景,为提升无监督事件抽取中触发词与论元抽取的稳定性与完整性,提出一种基于规则和统计学习的开放域事件抽取(RSL-ODEE)模型.基于事件语义学构建开放域事件模式(ES-ODES),将事件形式化为由触发词及主体、客体、时间、空间、工具、频度、度量等要素构成的结构化语义单元,以增强抽取过程的可解释性与跨领域泛化能力;在此模式指导下,构建融合触发词识别、抽象语义表示(AMR)语义图论元搜索与开放域机器阅读理解(MRC)论元矫正的无监督事件抽取模型,实现论元抽取、补全与断裂合并.结果表明:在中文突发事件语料库(CEC)事件提取数据集上,该方法在事件触发词与事件论元抽取的 F1 值分别仅比最优的有监督模型低 10.9%与 17.5%,具有可比较性与竞争力;在无监督条件下,引入事件语义学并结合 AMR 的结构性论元抽取与 MRC 的补全纠错能力,可有效缓解论元缺失与断裂问题,从而提升开放域事件抽取效果.

To improve the stability and completeness of trigger word and argument extraction in unsupervised event extraction for data scenarios with open domains and lack of task specific annotations,a rule and statisti-cal learning-based open domain event extraction model(RSL-ODEE)is proposed.Based on event semantics,an open domain event schema(ES-ODES)is proposed to formalize events as structured semantic units that com-posed of trigger words and elements such as subject,object,time,location,instrument,frequency,and meas-urement,thereby enhancing the interpretability and cross domain generalization ability of the extraction process.Under the guidance of this schema,an unsupervised event extraction framework is constructed that integrates trigger word recognition,abstract meaning representation(AMR)semantic graph-based argument search,and open domain machine reading comprehension(MRC)argument correction,to achieve argument extraction,completion,and fragment merging.Experimental results on the Chinese Emergency Corpus(CEC)event extraction dataset show that the proposed method achieves F1 scores for event trigger words and event argument extraction within 10.9%and 17.5%of the optimal supervised models,respectively,demonstrating comparability and competitiveness.Under unsupervised conditions,the incorporation of event semantics com-bined with AMR-based structural argument extraction and MRC-based completion and error correction capabil-ities can effectively alleviates argument omission and fragmentation issues,thereby improving the effectiveness of open domain event extraction.

尹积余;许晶晶;王兆阳;王华珍;周浩

华侨大学 计算机科学与技术学院,福建 厦门 361021华侨大学 计算机科学与技术学院,福建 厦门 361021华侨大学 计算机科学与技术学院,福建 厦门 361021华侨大学 计算机科学与技术学院,福建 厦门 361021华侨大学 计算机科学与技术学院,福建 厦门 361021

信息技术与安全科学

事件语义学开放域事件抽取抽象语义表示无监督

event semanticsopen domainevent extractionabstract meaning representationunsupervised

《华侨大学学报(自然科学版)》 2026 (5)

598-608,11

华侨大学中央高校基本科研业务费资助项目(2024HQYJ01)

10.11830/ISSN.1000-5013.202510064

评论