Proceedings of the Rapid Screening Technologies, Deception Detection and Credibility Assessment Symposium

This paper describes a new experimental system, the Automated Deception Analysis Machine (ADAM), that was used to conduct automated, chat-based interviews and perform deception detection of the text-based responses. 108 participants were interviewed by this new chatbot agent and they provided 3,162 chat-based responses. Several new metrics were discovered that highly correlate to deceit. ADAM captured extremely accurate response time measurements, clandestinely monitored the number of times the message was edited, and used linguistic features like lexical diversity and word count to analyze the replies. Response time, number of edits, and lexical diversity all were highly correlated to deceptive responses. The paper also describes several classification algorithms that were implemented by the system to discriminate between truthful and deceitful responses. The algorithms achieved a classification of over 90% in some instances for individual questions.