A framework to support intelligibility in pervasive applications

Adaptations of context-aware applications do not always result in behaviours that users expect, due to imperfect sensing of context information and variability in human preferences, etc. This can negatively impact the user experience of applications and compromise the trust users have in them. In order to gain user acceptance it is critical for applications to support intelligibility, so they are capable of justifying their adaptive actions and explaining the decision process of adaptations to their users. Based on these intelligible explanations, users should be able to modify application settings/thresholds to correct any undesirable behaviour. This paper presents a model-based developmental framework that supports intelligibility and user control of context-aware applications. It identifies and exposes the internal middleware models which influence adaptation decisions, and facilitates generations of explanations regarding evaluations of the models. These middleware models include preference models defined using Defeasible Logic, situation abstractions specified using Hidden Markov Models and First Order Logic, and context models developed using Context Modelling Language. The framework also takes into account users' expertise in technology when providing explanations and control to application behaviours.