Effects of different interaction contexts when evaluating gaze models in HRI