ParallelGAT:网络谣言检测方法

ParallelGAT:Network Rumor Detection Method

[研究目的]社交媒体平台在促进多元信息交互的同时,也助推了谣言的快速传播。如何准确、及时地发现谣言,已成为多领域学者共同关注的热点问题。最新的谣言检测研究表明,基于谣言传播结构的方法能够捕捉丰富的谣言传播特征,提升谣言检测准确率,而基于外部证据推理的方法可以在传播数据不充分的情况下判别谣言真假,提高谣言检测的时效性。[研究方法]为实现谣言检测准确率和时效性的同步提升,本研究结合这两种方法的优势,提出了基于并行图注意力网络的谣言检测方法ParallelGAT。ParallelGAT由两个图注意力网络模型BiGAT和MlGAT并行构成。其中,BiGAT模型通过引入注意力机制以捕捉重要的谣言传播和散布特征;MlGAT模型通过在外部知识中增加多头注意力机制以获取关键的外部句子级证据知识和词语级证据知识;BiGAT和MlGAT的输出特征向量最终通过聚合模块生成谣言检测标签。[研究结论]在公开数据集上的实验结果显示,该文提出的模型优于现有的方法。

[Research purpose]Social media platforms not only help the interaction of multiple information,but also incur the rapid spread of rumors.How to detect rumors accurately and timely has become a hot issue in many fields.The latest research on rumor detection shows that the method based on rumor propagation structure can capture rich rumor propagation characteristics and improve the accuracy of rumor detection,and the method based on external evidence reasoning can distinguish the truth and rumor in the case of insufficient propagation data and improve the timeliness of rumor detection.[Research method]In order to improve the accuracy and timeliness of rumor detection synchronously,this paper proposed a rumor detection method based on parallel graph attention network,called ParallelGAT.ParallelGAT is consisted of two graph attention network models in parallel,called BiGAT and MlGAT separately.BiGAT uses attention mechanism to capture the important rumor propagation and dissemination characteristics.MlGAT adds multi head attention mechanism to obtain key external sentence level evidence knowledge and word level evidence knowledge of external knowledge.The output eigenvectors of BiGAT and MlGAT finally generate rumor detection tags through the aggregation module.[Research conclusion]Experimental results on public open datasets show that the proposed model is superior to the existing methods.