Novel features for capturing cooccurrence behavior in dyadic collaborative problem solving tasks

Temporal patterns of verbal and non-verbal behavior in human interactions reveal our attitudes toward each other and can have an outsized impact on outcome of activities such as negotiations, coordination and team work. However, not much work has been done towards automatically analyzing the temporal aspects of such behavior during collaborative tasks. In this paper, we take an initial step toward this by proposing a novel feature that captures convergent or divergent behavior between dyads involved in a collaborative problem solving task. This feature, dubbed histograms of cooccurrence, capture how often different prototypical behavioral states exhibited by one person co-occur with those exhibited by his/her partner over different temporal lags. We show that not only does such a feature bring out the differences between dyads and nondyads, but is also interpretable in that it tells us which behavioral states are most likely to occur in true dyads as opposed to nominal or artificial dyads.