• •    

CCML2021+82: 基于异构图注意力网络的微博谣言监测模型MicroBlog-HAN

毕蓓1,2,潘慧瑶2,陈峰2,隋京言3,高扬4,王耀君2   

  1. 1. 北京理工大学计算机学院
    2. 中国农业大学信息与电气工程学院
    3. 中国科学院计算技术研究所
    4. 北京工业大学经济与管理学院
  • 收稿日期:2021-06-09 修回日期:2021-06-23 发布日期:2021-06-23
  • 通讯作者: 王耀君

CCML2021+82: MicroBlog Rumor Detection Model Based on Heterogeneous Graph Attention Network MicroBlog-HAN

  • Received:2021-06-09 Revised:2021-06-23 Online:2021-06-23

摘要: 社交媒体方便了人们的日常交流和信息传播,同时也是谣言滋生和传播的温床,因此如何在谣言传播早期自动监测极具现实意义。现有的检测方法没有充分利用微博信息传播图的语义信息,本研究基于异构图注意力网络(HAN)构建了谣言监测模型MicroBlog-HAN。该模型采用分层注意力机制:节点级注意力和语义级注意力。首先,节点级注意力机制结合微博节点的邻居,生成两组具有特定语义的节点嵌入。然后,语义级注意力再融合不同语义,得到最终的节点嵌入,输入到分类器中执行二分类任务。最后,给出输入微博是谣言还是非谣言的分类结果。在两个真实的微博谣言数据集上的实验结果表明MicroBlog-HAN模型可以实现微博谣言较准确的识别,准确率为87%以上。

关键词: 微博, 谣言监测, 异构图, 元路径, 异构图注意力网络

Abstract: Social media highly facilitates people's daily communication and disseminating information, but unfortunately, it is also an ideal hotbed for the breeding and dissemination of Internet rumors. As a result, How to automatically monitor rumor dissemination in the early stage is of great practical significance. To date, the existing detection methods have all failed to take full advantage of the semantic information of the microblog information propagation graph. In this study, based on a heterogeneous graph attention network (HAN), we built a rumor monitoring model, namely MicroBlog-HAN. The model makes use of a hierarchical attention mechanism: node-level attention and semantic-level attention. First of all, the node-level attention mechanism combines with the neighbors of microblog nodes to generate two groups of node embeddings with specific semantics. After that, semantic-level attention fuses different semantics to get the final node embedding of microblog, which is then treated as the classifier's input to perform the two classification task. In the end, the classification results of whether the microblog is rumor or non-rumor are given. The experimental results on two real-world microblog rumor datasets convincingly prove that the Microblog-Han model can accurately identify microblog rumors, with an accuracy rate of over 87%.

Key words: microblog, rumor detection, heterogeneous graph, meta-path, heterogeneous graph attention network

中图分类号: