Abstract
Users of social networks tend to post and share content with little restraint. Hence, rumors and fake news can quickly spread on a huge scale. This may pose a threat to the credibility of social media and can cause serious consequences in real life. Therefore, the task of rumor detection and verification has become extremely important. Assessing the veracity of a social media message (e.g., by fact checkers) involves analyzing the text of the message, its context and any multimedia attachment. This is a very time-consuming task that can be much helped by machine learning. In the literature, most message veracity verification methods only exploit textual contents and metadata. Very few take both textual and visual contents, and more particularly images, into account. In this paper, we second the hypothesis that exploiting all of the components of a social media post enhances the accuracy of veracity detection. To further the state of the art, we first propose using a set of advanced image features that are inspired from the field of image quality assessment, which effectively contributes to rumor detection. These metrics are good indicators for the detection of fake images, even for those generated by advanced techniques like generative adversarial networks (GANs). Then, we introduce the Multimodal fusiON framework to assess message veracIty in social neTwORks (MONITOR), which exploits all message features (i.e., text, social context, and image features) by supervised machine learning. Such algorithms provide interpretability and explainability in the decisions taken, which we believe is particularly important in the context of rumor verification. Experimental results show that MONITOR can detect rumors with an accuracy of 96% and 89% on the MediaEval benchmark and the FakeNewsNet dataset, respectively. These results are significantly better than those of state-of-the-art machine learning baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2109.02271
