Combining CSP and Constraint-based Mining for Pattern Discovery

Medhi Khiari (1) , Patrice Boizumault (1) , Bruno Crémilleux (1)

1 Equipe CODAG - Laboratoire GREYC - UMR6072

A well-known limitation of a lot of data mining methods is the huge number of patterns which are discovered: these large outputs hamper the individual and global analysis performed by the end-users of data. That is why discovering patterns of higher level is an active research field. In this paper, we investigate the relationship between local constraint-based mining and constraint satisfaction problems and we propose an approach to model and mine patterns combining several local patterns, i.e., patterns defined by n-ary constraints. The user specifies a set of n-ary constraints and a constraint solver generates the whole set of solutions. Our approach takes benefit from the recent progress on mining local patterns by pushing with a solver on local patterns all local constraints which can be inferred from the n-ary ones. This approach enables us to model in a flexible way any set of constraints combining several local patterns. Experiments show the feasibility of our approach

Domaines: Intelligence artificielle [cs.AI]; Traitement du texte et du document; Apprentissage [cs.LG]

Medhi Khiari, Patrice Boizumault, Bruno Crémilleux. Combining CSP and Constraint-based Mining for Pattern Discovery. Advances in Knowledge Discovery and Management 2 (Post-EGC'10 Selected Papers), Studies in Computational Intelligence, Vol. 398, Springer, pp.85-104, 2012. ⟨hal-01022438⟩