Towards Measuring Human Interactions in Conversational Settings

Abstract

The Facilitator Room is a conference room that is outfitted with sensors and actuators in order to observe and influence human behavior in conversational settings. This work describes our efforts thus far in developing robust sensing mechanisms in the visual and auditory domains and designing statistical models to analyze and predict behavior. We review the “Influence Model,” our primary analysis tool, which we developed for this purpose in [1]. The “Influence Model” models a group of interacting agents as a group of simple Markov chains that are each influencing each other’s state transitions. We demonstrate the capabilities of this model on both synthetic data and real interaction data from the Facilitator Room. We describe our approaches for doing prediction with this model and close with a discussion of how we plan to influence interactions with the room’s actuators.