CATCH ME IF YOU CAN: DETECTING ONLINE DECEPTION USING AUTOMATED LINGUISTIC ANALYSIS

Abstract

Despite the widespread problem of online falsehoods in social media, research in deception on these platforms remains limited. Using linguistic analysis methods, this research aims to investigate linguistic properties of deception across private and public settings in two studies. Study 1 explores the linguistic differences between truths and lies written across both settings, and the effect of platform privacy on the linguistic properties of messages. Ninety-eight participants were recruited to complete an online task requiring them to write three truths and three lies. Participants were either told that their responses would be collected for private analysis, or to be made viewable to others on Twitter. Results show a within-effect for linguistic differences between truths and lies in both conditions. Between-effects were found when truths or lies written in the private condition were compared to those in the public condition. Study 2 compared the accuracy of deception detection between human judges and an automated model. Results show that human judges performed significantly better than the model for samples from the public condition. Implications of findings are discussed in relation to the use of linguistic-based cues to detect deception in private settings and on social media.