Estimation of Empathy Skill Level and Personal Traits Using Gaze Behavior and Dialogue Act During Turn-Changing

We explored the gaze behavior towards the end of utterances and dialogue acts (DAs), i.e., verbal-behavior information indicating the intension of an utterance, during turn-keeping/changing to estimate several social skills and personal traits in multi-party discussions. We first collected data on several personal indicators, i.e., Big Five, which measures personal traits, and Davis’ Interpersonal Reactivity Index (IRI), which measures empathy skill level, utterances that include DA categories, and gaze behavior, from participants in four-person discussions. We constructed and evaluated models for estimating the scores of these indicators using gaze behavior and DA information. The evaluation results indicate that using both gaze behavior and DAs during turn-keeping/changing is effective for estimating all such scores with high accuracy. It is also possible to estimate these scores with higher accuracy by using the gaze distribution to the current speaker and listener and amount of speaking obtained during the entire discussion. We also found that the IRI scores can be estimated more accurately than those of Big Five.