The researchers examine the use of linguistic features to detect deception in interview dialogues, with the goal of improving classification performance beyond human ability. Their analysis shows that there are significant differences in language use between truthful and deceptive responses to interview questions, and that there are variations in deception patterns based on gender and native language.
They use these findings to select features for machine learning experiments, achieving a 72.74 F1-Score, about 27% better than human performance, by combining linguistic features and individual traits. Their study highlights the potential of using linguistic features and machine learning to improve deception detection in interview settings.
Read the research: Linguistic Cues to Deception and Perceived Deception in Interview Dialogues