I Know What You Want: Using Gaze Metrics to Predict Personal Interest

Abstract

In daily communications, we often use interpersonal cues - telltale facial expressions and body language - to moderate responses to our conversation partners. While we are able to interpret gaze as a sign of interest or reluctance, conventional user interfaces do not yet possess this possible benefit. In our work, we evaluate to what degree fixation-based gaze metrics can be used to infer a user's personal interest in the displayed content. We report on a study (N=18) where participants were presented with a grid array of different images, whilst being recorded for gaze behavior. Our system calculated a ranking for shown images based on gaze metrics. We found that all metrics are effective indicators of the participants' interest by analyzing their agreement with regard to the system's ranking. In an evaluation in a museum, we found that this translates to in-the-wild scenarios despite environmental constraints, such as limited data accuracy.