时序感知的异质图神经谣言检测  

Sequence-aware Heterogeneous Graph Neural Rumor Detection

作　　者：陈林威 宋玉蓉 宋波[2] CHEN Linwei;SONG Yurong;SONG Bo(College of Automation&College of Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;School of Modern Posts&Institute of Modern Posts,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学自动化学院、人工智能学院,南京210023 [2]南京邮电大学现代邮政学院,南京210003

出　　处：《小型微型计算机系统》2024年第1期45-51,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61672298,61873326)资助;江苏高校哲学社会科学研究重点项目(2018SJZDI142)资助;江苏省高等学校自然科学研究面上项目(20KJB120007)资助;江苏省自然科学基金青年基金项目(BK20200758)资助.

摘　　要：近年来,在线社交媒体的发展大大加速了谣言的滋生和传播,谣言的危害性使得谣言的自动检测技术受到研究学者的广泛关注.本文同时考虑事件与事件之间的全局结构关系以及事件内部消息传播的时序关系,以异质图为载体共同显式建模两种关系,提出一种新的时序感知的异质图神经谣言检测模型.该模型利用时序感知的自注意力机制捕获事件内部转发(或评论)贴之间的时序关系,并将具有时序信息的转发(或评论)贴与源贴融合,得到事件的局部时序表征;接着利用元素级注意力机制捕捉事件与事件之间的全局结构关系,学习事件的全局结构表征;最后将二者融合用于检测谣言.实验结果表明,该模型优于大多数现有模型,可以提高谣言检测性能,并且同样具有优秀的早期检测性能.In recent years,the development of online social media has greatly accelerated the breeding and spreading of rumors,and the harmful effects of rumors have led to the automatic detection of rumors receiving extensive attention from research scholars.In this paper,we propose a new sequence-aware heterogeneous graph neural(SHGN)rumor detection model by considering both the global structural relation between events and events and the temporal relation of message propagation within events,and modeling the two relations together explicitly with a heterogeneous graph as a carrier.The model captures the temporal relations between retweets(or comment)posts within an event using a temporal-aware self-attention mechanism,and fuses the retweets(or comment)posts with temporal information with the source posts to obtain the local temporal representation of the event;then captures the global structural relationships between the event and the event using an element-level attention mechanism to learn the global structural representation of the event;finally,the two are fused for detecting rumors.Experimental results show that the model outperforms most existing models to improve rumor detection performance and also has excellent early detection performance.

关 键 词：时序感知 异质图 谣言检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]