Saint or Sinner? Language-Action Cues for Modeling Deception Using Support Vector Machines

Abstract

In text-based online communication, the clues available to the communicator for ascertaining the underlying intent of a message sender and discerning whether a message is deceptive are often limited to the text. Nonetheless, research has shown that it is possible to detect deception with reasonable accuracy by applying certain classification methodologies to certain observable language-action cues. This paper explores the viability of adopting support vector machines (SVMs) to develop an automated process for deception detection in computer-mediated communications (CMC). In particular, it examines the prediction accuracy of SVM models with different kernel functions on data collected from a controlled online interactive game set up on a Google+ Hangout platform. The results indicate that SVM models using the radial basis function (RBF) kernel can classify the complex relationships with high accuracy between language-action cues and deception.

Keywords: language-action cues, computer-mediated deception, online game experiments