Propagation-Based Fake News Detection Using a Combination of Different Content Features
Propagation-Based Fake News Detection Using a Combination of Different Content Features
复制标题
DOI：
10.1109/gcce56475.2022.10014073
复制
发表时间：
2022-10
期刊：
影响因子：
--
通讯作者：
Kayato Soga;Soh Yoshida;M. Muneyasu
中科院分区：
文献类型：
--
作者：
Kayato Soga;Soh Yoshida;M. Muneyasu
虚假新闻检测是安全使用社交网络服务的一项紧迫任务，这些服务的可信度无法验证。已经提出了使用从新闻文章中提取的基于风格的特征的基于内容的方法和关注假新闻和真实的新闻（其不包含虚假信息）之间的不同传播模式的基于上下文的方法。最近，人们对通过将图神经网络应用于表示用户交互的图来提取基于传播的特征的方法很感兴趣，其中基于风格的特征被分配为初始状态。然而，现有方法存在网络训练后鲁棒性不足的问题。本文提出了一种基于传播特征提取的假新闻检测方法，该方法结合了不同的基于风格的特征，以提高基于传播特征提取的性能。使用Twitter数据进行的比较实验证实了所提出的方法的有效性。
Fake news detection is an urgent task for the safe use of social networking services whose credibility cannot be verified. Content-based methods using style-based features extracted from news articles and context-based methods focusing on the different propagation patterns between fake news and real news (which does not contain false information) have been proposed. Recently, there has been interest in methods that extract propagation-based features by applying graph neural networks to a graph representing user interactions, with style-based features assigned as initial states. However, existing methods have the problem of insufficient robustness after a network is trained. This paper proposes a fake news detection method that combines different style-based features to improve the performance of propagation-based feature extraction. Comparison experiments conducted using Twitter data confirm the effectiveness of proposed method.