Un algorithme générique d’extraction de bi-ensembles sous contraintes dans des données booléennes

The availability of complete algorithms for constraint-based mining of formal concepts (e.g., frequent ones) enables to process large boolean datasets and can be used in many application domains. Considering noisy data, it seems important to extend formal concepts towards fault-tolerance. We embed formal concept extraction into a more general framework for constraint-based mining of bi-sets, i.e., computing associations of sets of objects with sets of attributes that satisfy some constraints. We propose an original, correct and complete algorithm for constraint-based bi-set mining. Its genericity supports the discussion of fundamental mechanisms like the declarative specification of the pattern properties, the enumeration and the constraint propagation strategies. Two concrete instances of this algorithm are presented and evaluated. Key-words: Data mining, formal concepts, fault-tolerance