Deducing evidence for social situations from dynamic geometric interaction data

We discuss how time-independent and time-dependent features of human social interaction geometry on small temporal and spatial scales may be used to extract evidence for or against the existence of social situations as a simple form of social context. Aside from providing a new method for quantitative investigation of human interaction behaviour, the ultimate vision motivating this research focuses on mobile devices autonomously measuring and processing data on interaction geometries in order to derive social situation context that can be used in mobile social networking scenarios. Our method is tested via an experiment using an IR tracking method already allowing for the precise determination of interpersonal distances and relative body orientation in a conversational setting. We investigate the performance of time-independent classifiers for the prediction of the involvement of pairs of persons in a social situation using relative distance and orientation. We then discuss results of using HMMs for exploiting the time-dependency of the interaction geometry.