These days, people have increasingly used social media as a go-to resource for any information need and daily news diet. In the past decade, the news ecosystem and information flow have been dramatically transformed by the popularity of such platforms. Social media users can, in fact, easily access nearly any kind of information and then spread it nearly without friction through activities like tweets/retweets in Twitter (now X) and similar means on other social media. This seemingly innocuous activity of spreading information has a collective consequence of making social media users responsible for radical changes in the way news is distributed, including both authentic and fake news. Moreover, malicious individuals have been implicated in capitalizing on the ease of introducing and spreading information in these platforms to create misinformation, spread it to a wider audience, and subsequently influence public opinion on important topics through information diffusion. Therefore, understanding the factors that motivate a user’s decision to share is of paramount importance in understanding the information diffusion phenomenon in social media.
In this paper, we propose an approach based on the Diffusion of Innovation theory to model, characterize, and compare real and fake news sharing in social media with a focus on different levels of influencing factors including innovation, communication channels, and social system. We apply that approach to identify factors related to the spread of fake news as they relate to users, the structure of news items themselves, and the networks through which news is circulated. We address the problem of predicting real and fake news sharing as a classification task and demonstrate the potentials of the proposed features by achieving an AUROC of around 0.97 and an average precision ranging from 0.88 to 0.95, consistently outperforming baseline models with a higher margin (at least 13% of average precision). In addition, we also found out that empirically identifiable characteristics of news items themselves and users who share news are the strongest element allowing accurate prediction of real and fake news sharing, followed by network-based features. Moreover, our proposed approach can be effectively used to model news diffusion as a multi-step propagation process.
