Towards Defining a Quality-Metric for Affective Feedback in an Intelligent Environment

Inference-based computational models are often prone to errors. For a Context-aware Intelligent Environment (CAIE), such errors in inference may result in contextual services that are not appropriately matched with a user’s current goals resulting in negative emotional states such as frustration for the user. Learning to infer such negative emotions from a user’s psychophysiological responses may enable the CAIE to deduce an implicit feedback signal regarding this appropriateness. Previously we have presented a framework for implementing this inference based affective feedback in a CAIE, including a method of ground truth data collection. We now propose a method for defining a quality metric on this inference derived from the underlying physiological processes, which is envisioned to enable us in defining a user-specific service-appropriateness feedback model. The results from the electrodermal activity (EDA) dataset collected from a Virtual-reality based Supermarket setup provide a preliminary validation to our EDA data based quality metric.