Lying trolls: Detecting deception and text-based disinformation using machine learning

This chapter explores how well-studied machine learning algorithms like Naive Bayes, Support Vector Machines, Decision Trees, Logistic Regression, and Random Forest, which have been successful in binary text classification for unsolicited emails especially spam can be adopted in detecting deception and text-based disinformation. This study explores how we can use psycholinguistic and computational linguistic processes linked to deception and cybercrime to generate models for detecting deception and text-based disinformation. The study attempts to establish whether there is a similarity in patterns of deception between text-based disinformation and other forms of text-based cybercrime like scams, deceptive online reviews, and fraud. We generated singleton and hybrid models to detect deception and text-based disinformation. The singleton models are trained on fraud, scams, favorable online reviews, and unfavorable online reviews. On the other hand, to generate the hybrid models we combined a disinformation train set to each of the singleton models. When evaluating our models we only considered models whose accuracies for detecting both deceptive and truthful instances were over 50%. Comparatively, there were more hybrid models which detected deception and text-based disinformation with predictive accuracies ranging from 60% to 100%.