Combining Evidence for Social Situation Detection

Social Situations (SSs) are social context models indicating nary social interaction on small spatio-temporal scales detected by means of Social Signal Processing (SSP). We discuss the problem of how to combine evidence from several sensor-sources for the benefit of algorithmic assessment of SS in a distributed agent-based Social Networking scenario. We propose a solution based on Subjective Logic (SL) that mediates between exchanging & processing of (a) raw low level sensor data, of (b) intermediate results of ´sub-symbolic´ probabilistic models typically used for SSP, and of (c) the final ´symbolic´ SS models. We evaluate key aspects of the approach on the basis of a social experiment, combining audio-based and geometry-of-interaction-based methods for SS detection.