基于多传递影响力的社交媒体谣言检测方法

社交媒体谣言检测是当前研究的热点问题，现有方法多数通过获取大量用户属性学习用户特征，但不适用于谣言的早期检测，忽略了用户之间的潜在关系对信息传播的影响。提出一种基于多传递影响力的谣言检测方法，根据源微博及其对应转发（评论）之间的关系构建文本信息传播图，并通过图卷积神经网络来捕获、学习文本信息的传播特征。利用文本信息和用户传播过程中的影响力，丰富可用于谣言检测早期的检测信息。将存在转发关系的用户构成用户影响力传播图，构建一种用户节点影响力学习方法，获取用户节点影响力，以增强用户特征信息。在此基础上，将文本特征与用户特征融合以进行谣言检测，从而提升检测效果。在3个真实社交媒体数据集上的实验结果表明，该方法在谣言自动检测以及早期检测的效果都有显著提升，与目前最好的基准方法相比，在微博、Twitter15、Twitter16数据集上的正确率分别提高了2.8%、6.9%和3.4%。

Social Media Rumor Detection Method Based on Multi-Transmit Influence

Social media rumor detection is a hot topic in current research.Existing methods learn user characteristics by acquiring many user attributes.However, they ignore the influence of potential relationships between users on information propagation, making them inapt for early detection of rumors.This paper proposes a Multi-Transmit Influence(MTI) model for social media rumor detection.The forwarding relationship between the source microblog and its corresponding forwards(comments) is used to construct a text information propagation graph.A graph convolution neural network is used to capture and learn the propagation characteristics of text information.In addition, the influence of users in the communication process is integrated to enrich information detection for early rumor detection.First, the users with forwarding relationships are formed into a user influence propagation graph.A user-node influence learning method is then constructed to capture the user-node influence thatenhances the user characteristic information.Finally, text and user features are fused to detect rumors more accurately.Experiments on three real social media data sets revealthat the proposed method significantly improved automatic and early rumor detection.Compared with the conventional benchmark methods, the accuracy improved by 2.8%, 6.9%, and 3.4% when tested using Weibo, Twitter15, and Twitter16 data sets, respectively.