Knowledge Discovery from Qualitative Spatial and Temporal Data

Qualitative reasoning formalisms facilitate the representation and interpretation of information involving complex entities. We use in this paper qualitative spatial and temporal reasoning to introduce novel data mining tasks, which consist in extracting knowledge from quantitative databases that are transformed into collections of qualitative relation networks (QRNs). After describing our qualitative data mining framework, we first propose an Apriori-like algorithm that exploits monotonicity and QRN consistency for pruning the search space: the validity of a pattern candidate depends on the supports of the larger patterns that include it and on its consistency. We then introduce an encoding of our data mining tasks into the well-known problem of frequent itemset mining. We finally show the feasibility of our approach by providing preliminary experimental results using real-world datasets about the movements of football players during matches.