联合异质图卷积网络和注意力机制的假新闻检测  

Joint Heterogeneous Graph Convolutional Network and Attention Mechanism for Fake News Detection

作　　者：韩晓鸿[1] 赵梦凡 张钰涛 HAN Xiaohong;ZHAO Mengfan;ZHANG Yutao(School of Information&Electrical Engineering,Hebei University of Engineering,Handan 056038,China;College of Intelligence and Computing,Tianjin University,Tianjin 300350,China)

机构地区：[1]河北工程大学信息与电气工程学院,河北邯郸056038 [2]天津大学智能与计算学部,天津300350

出　　处：《小型微型计算机系统》2024年第2期301-308,共8页Journal of Chinese Computer Systems

基　　金：河北省自然科学基金项目(F2020402003)资助。

摘　　要：社交媒体平台的开放性和包容性为人们提供了自由的表达方式,但也引发了新的社会问题,假新闻在社交平台层出不穷,会引起公众恐慌,侵害人们的精神健康,这使得假新闻检测尤为必要.现有的假新闻检测方法大多侧重于从文本内容、用户和传播模式中挖掘有效信息,但是这些方法没有充分利用文本内容的全局语义关系.为了有效融合新闻内容的全局语义信息和新闻传播的全局结构关系,本文提出一种基于元路径的推文-词-用户异质图卷积注意力框架HGCAN,根据元路径将构建的推文-词-用户异质图分解为两个子图,通过图卷积网络提取传播结构特征,利用注意力机制聚合邻居节点的信息并学习子图重要性,从而有效学习节点的特征表示.在两个公开数据集上的实验结果表明,相比于其他方法,本文方法在准确率和F1指标上都取得了较为先进的结果.The openness and inclusiveness of social media platforms provide people with free expression,but also cause new social problems.Fake news emerges endlessly on social platforms,causing public panic and jeopardizing people′s mental health,which makes fake news detection all the more necessary.Most of the existing fake news detection methods focus on mining effective information from text content,users and propagation patterns,but these methods do not fully exploit the global semantic relationship of text content.In order to effectively integrate the global semantic information of news content and the global structural relationship of news dissemination,this paper proposes a meta-path-based twitter-word-user heterogeneous graph convolutional attention framework HGCAN.The constructed heterogeneous graph is decomposed into two subgraphs according to the meta-path.Extracting Text Content Features via graph convolutional networks,and aggregating the information of neighbor nodes and learning the subgraph importance via attention mechanism,so as to effectively learn the feature representation of nodes.Experimental results on two public datasets show that,compared with other methods,our method achieves relatively advanced results in both accuracy and F1 metrics.

关 键 词：异质图 图卷积网络 注意力机制 假新闻检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]