Abstract: PURPOSE: It is difficult to simulate real deception in laboratory studies. Therefore, some researchers proposed to simulate the emotional arousal state of the real situation by inducing a high emotional arousal of the subjects before the experiment, so as to increase the ecological validity of the laboratory research results. Research has shown that cheating activates both happiness and fear more. This study will explore whether pleasurable emotion eliculation causes cheaters to expose more facial deception cues in a laboratory setting.
METHODS: Using the Guilty Knowledge Tests, the facial expressions of the emotion-induced group and the neutral emotion-induced group were used as the material for the negative response to the detection stimulus (i.e. the deception response). We obtained 47,097 frames of deception material in the emotion-induced group and 49,529 frames of deception material in the neutral emotion group (frame rate 50 f/s). OpenFace analysis using computer vision software subjects facial AU frequency, and each frame material related to deception in AU unit (AU01 AU02, AU04, AU05, AU06, AU07, AU12, AU20, AU26) intensity input machine learning software WEKA.
RESULTS: The deception AU activation frequency of emotion-induced group and neutral emotion-induced group was significantly different on AU12 (t = 2.470, P = 0.027, effect size = 0.638). Based on the AU intensity related to deception, the machine learning classification of whether the material is emotionally induced is performed, and the accuracy of the three classifiers is higher than 95%.
CONCLUSIONS: Inducing the participants' happiness before the experiment can make the participants reveal more facial deception cues when they cheat.

Key words: emotion elicitation, deception cue, computer vision, machine learning