Prediction and explanation over DL-Lite data streams

Abstract

Stream reasoning is an emerging research area focusing on the development of reasoning techniques applicable to streams of rapidly changing, semantically enhanced data. In this paper, we consider data represented in Description Logics from the popular DL-Lite family, and study the logic foundations of prediction and explanation over DL-Lite data streams, i.e., reasoning from finite segments of streaming data to conjectures about the content of the streams in the future or in the past. We propose a novel formalization of the problem based on temporal \past-future" rules, grounded in Temporal Query Language. Such rules can naturally accommodate complex data association patterns, which are typically discovered through data mining processes, with logical and temporal constraints of varying expressiveness. Further, we analyse the computational complexity of reasoning with rules expressed in different fragments of the temporal language. As a result, we draw precise demarcation lines between NP-, DP- and PSpace-complete variants of our setting and, consequently, suggest relevant restrictions rendering prediction and explanation more feasible in practice.