Estimating the Tendency of Social Media Users to Spread Fake News

The unique characteristics of social media, such as popularity, ubiquitousness, and inadequate supervision, make it a perfect medium for fake news propagation. While users play a critical role in this propagation, not all of them have the same level of impact and involvement. Identifying the news-sharing behaviors of different users and predicting them automatically can be a leading step toward detecting fake news and understanding the factors that contribute to its spread. Previous attempts to detect fake news spreaders have focused on binary classification, assuming users as either spreaders or non-spreaders of fake news. To address this oversimplification, we propose estimating users’ tendency to spread fake news by introducing a metric that represents the degree of users’ propensity to spread misinformation. Our provided approach is a supervised regression model utilizing text-based features extracted from users’ writings on social media. We created and annotated a new dataset based on FakeNewsNet, a popular data repository on fake news detection, to train our model and conduct our experiments. In our experiments, we establish the practicality of our approach by achieving a Root Mean Squared Error (RMSE) of 0.26, using a range of values from 0 to 1 to represent users’ inclination to spread fake news. We also demonstrate that utilizing text-based features leads to better performance than using explicit features directly provided by social media.