Concurrent Inference on High Level Context Using Alternative Context Construction Trees

Context aware services rely on context information. However, in many cases this information is not available. In this case, the missing context information can often either be inferred from existing knowledge using different inference approaches, or accessed from external context providers. Thereby, dealing with uncertainty is another major concern. In this paper we present a generic approach called alternative context construction trees (ACCTs) that enables the concurrent evaluation and consolidation of different alternative inference approaches, like logic rules, Bayesian networks and CoCoGraphs. In particular, we introduce a common signature structure to make those alternatives comparable in a generic way. Thereby, the approach is able to adapt dynamically to specific service requirements with respect to the accuracy of available context information. In addition, we present so called Bayesian network templates that enable a light-weight inference of high-level context information if only probabilistic knowledge about the causal interdependencies of context artifacts is available.