Compact Representation of Conditional Probability for Rule-Based Mobile Context-Aware Systems

Context-aware systems gained huge popularity in recent years due to rapid evolution of personal mobile devices. Equipped with variety of sensors, such devices are sources of a lot of valuable information that allows the system to act in an intelligent way. However, the certainty and presence of this information may depend on many factors like measurement accuracy or sensor availability. Such a dynamic nature of information may cause the system not to work properly or not to work at all. To allow for robustness of the context-aware system an uncertainty handling mechanism should be provided with it. Several approaches were developed to solve uncertainty in context knowledge bases, including probabilistic reasoning, fuzzy logic, or certainty factors. In this paper, we present a representation method that combines strengths of rules based on the attributive logic and Bayesian networks. Such a combination allows efficiently encode conditional probability distribution of random variables into a reasoning structure called XTT2. This provides a method for building hybrid context-aware systems that allows for robust inference in uncertain knowledge bases.