Susceptibility of Communities Against Low-Credibility Content in Social News Websites

Social news websites, such as Reddit, have evolved into prominent platformsfor sharing and discussing news. A key issue on social news websites sites isthe formation of echo chambers, which often lead to the spread of highly biasedor uncredible news. We develop a method to identify communities within a socialnews website that are prone to uncredible or highly biased news. We employ auser embedding pipeline that detects user communities based on their stancestowards posts and news sources. We then project each community onto acredibility-bias space and analyze the distributional characteristics of eachprojected community to identify those that have a high risk of adopting beliefswith low credibility or high bias. This approach also enables the prediction ofindividual users' susceptibility to low credibility content, based on theircommunity affiliation. Our experiments show that latent space clusterseffectively indicate the credibility and bias levels of their users, withsignificant differences observed across clusters – a 34% difference in theusers' susceptibility to low-credibility content and a 8.3% difference inthe users' susceptibility to high political bias.