Towards context-adaptive affective computing

Affective computing systems with situated or ambient intelligence could be extremely effective in various application scenarios. However, majority of the proposed systems have limited utility since either they are strictly context-sensitive or otherwise too general. In this paper, we report on building and evaluating a context-aware yet situation-adaptive Bayesian inference framework that predicts human mental states in varying contexts. We use real and synthetic data to validate our hypothesis by modeling a three-layered Bayesian network (BN). We test this BN with two, and three context scenario, incrementally. The network gives an accuracy of above 85%. Thus, the framework may be utilized for multiple contexts, in variety of affective computing application scenarios.