News on Twitter: Engagement, Exposure and Estimating Credibility using Machine Learning

Abstract

Online health-related information is increasing rapidly, and social media is a favorite medium for its dissemination to reach a wider audience. Users must know the authenticity and credibility of this information, but manually evaluating them is resource-intensive; therefore, developing methods to automatically estimate the credibility of health-related information shared on social media is increasingly important. This study was explored to develop machine learning-based methods to estimate the credibility of health-related information on social media. Data were obtained from a public repository that consists of tweets and health-related web pages shared through those tweets on social media. The web pages were manually labeled against 10-point checklist criteria and were used to train and test machine learning models for their ability to estimate these 10-point checklist criteria automatically. The best performing machine learning model was the XG Boost model, with an average accuracy of 95%. The study also investigated the engagement of social media users on these web pages. It is found that low credibility web pages were shared more often on social media than high credibility web pages.