An empirical study of eye-gaze behaviors: towards the estimation of conversational engagement in human-agent communication

In face-to-face conversations, speakers are continuously checking whether the listener is engaged in the conversation by monitoring the partner's eye-gaze behaviors. In this study, focusing on eye-gaze as information of estimating user's conversational engagement, first, we conduct a Wizard-of-Oz experiment to collect the user's gaze behaviors as well as the user's subjective reports and an observer's judgment concerning the user's engagement in the conversation. Then, by analyzing the user's gaze behaviors, variables and factors for estimating the user's engagement are identified. Based on the analysis, we propose four types of engagement estimation methods based on gaze duration information and gaze transition 3-gram patterns. As the results of comparing the performance of these methods, it is revealed that a method which takes account of the individual differences in gaze transition patterns performs the best and can predict the user's conversational engagement quite well.