摘要
事件预测旨在结合事件的语义信息与结构关系,实现对未来事件的精准推断。针对现有图神经网络方法中存在语义捕捉不足及外部知识整合有限的问题,提出一种基于语义增强与候选排序优化的背景感知事件预测方法(SECRO)。该方法采用三阶段框架:首先,利用大语言模型生成高质量的事件节点嵌入,弥补语义表达的不足;其次,基于图神经网络建模事件间的结构与关联关系,生成初步预测结果;最后,设计了一种候选排序优化机制,结合大语言模型中蕴涵的世界知识提升事件预测精度。在三个公开数据集上的实验结果表明,该方法在平均排名分数(MRR)上分别较RGCN和SeCoGD方法提升了8.34和6.84个百分点,取得了新的SOTA性能。扩展实验结果进一步验证了该方法能够增强现有图方法在事件预测任务中的性能。
Event forecasting aims to integrate the event semantic information with structural relationships to precisely forecast future events.To address the issues of insufficient semantic capture and limited external knowledge integration in existing graph neural network methods,this paper proposed a context-aware event forecast method based on semantic enhancement and candidate ranking optimization(SECRO).The method employed a three-stage framework:firstly,a large language model generated high-quality event node embeddings to address semantic expression deficiencies;secondly,a graph neural network modeled the structural and relational connections among events,generating the preliminary prediction results;lastly,a candidate ranking optimization mechanism integrated world knowledge from the large language model,enhancing the event prediction accuracy.Experiments on three public datasets show that the method improves mean reciprocal rank(MRR)by 8.34 and 6.84 percentage points over RGCN and SeCoGD respectively,achieving state-of-the-art performance.Extended experimental results further confirm that the method enhances the performance of existing graph-based approaches for event prediction.
作者
马荣
马博
王震
艾孜麦提·艾尼瓦尔
杨雅婷
王磊
Ma Rong;Ma Bo;Wang Zhen;Azmat Anwar;Yang Yating;Wang Lei(Xinjiang Technical Institute of Physics&Chemistry,Chinese Academy of Sciences,rümqi 830011,China;University of Chinese Aca-demy of Sciences,Beijing 100049,China;Xinjiang Laboratory of Minority Speech&Language Information Processing,Chinese Academy of Sciences,rümqi 830011,China)
出处
《计算机应用研究》
北大核心
2025年第9期2599-2606,共8页
Application Research of Computers
基金
新疆维吾尔自治区“天山英才”科技创新领军人才资助项目(2022TSYCLJ0046)
新疆维吾尔自治区自然科学基金重点项目(2023D01D17)
新疆维吾尔自治区“天山英才”培养计划资助项目(2023TSYCCX0041,022TSYCCX0059)
中国科学院青年创新促进会优秀会员资助项目(Y2023118,Y2021112)
新疆维吾尔自治区重点研发任务专项资助项目(2023B03024)
新疆维吾尔自治区自然科学基金资助项目(2022D01B207)
新疆维吾尔自治区上海合作组织科技伙伴计划及国际科技合作计划资助项目(2023E01019)。
关键词
语义增强
事件预测
图神经网络
大语言模型
semantic enhancement
event forecasting
graph neural network
large language model
