Abstract

This study investigates how fake news shared on social media platforms can be automatically identified. Drawing on the Elaboration Likelihood Model and previous studies on information quality, we develop and test an explorative research model on Facebook news posts during the U.S. presidential election 2016. The study examines how cognitive, visual, affective and behavioral cues of the news posts as well as of the addressed user community can be used by machine learning classifiers to identify fake news fully automatically. The best performing configurations achieve a stratified 10-fold cross validated predictive accuracy of more than 80%, and a recall rate (share of correctly identified fake news) of nearly 90% on a balanced data sample solely based on data directly available on Facebook. Platform operators and users can draw on the results to identify fake news on social media platforms - either automatically or heuristically.

Download

Share

COinS