Constructing Iceberg Lattices from Frequent Closures Using Generators

Laszlo Szathmary (1) , Petko Valtchev (1) , Amedeo Napoli (2) , Robert Godin (1)

1 LATECE Laboratory - UQAM Montreal - Laboratory for Research on Technology for ECommerce
2 ORPAILLEUR - Knowledge representation, reasonning

Frequent closures (FCIs) and generators (FGs) as well as the precedence relation on FCIs are key components in the definition of a variety of association rule bases. Although their joint computation has been studied in concept analysis, no scalable algorithm exists for the task at present. We propose here to reverse a method from the latter field using a fundamental property of hypergraph theory. The goal is to extract the precedence relation from a more common mining output, i.e. closures and generators. The resulting order computation algorithm proves to be highly efficient, benefiting from peculiarities of generator families in typical mining datasets. Due to its genericity, the new algorithm fits an arbitrary FCI/FG-miner.

Mots clés: formal concept analysis; concept lattices; iceberg lattices; hypergraphs; transversal hypergraph; precedence relation; frequent closed itemsets; frequent generators; association rules; pattern mining

Domaines: Recherche d'information [cs.IR]

Laszlo Szathmary, Petko Valtchev, Amedeo Napoli, Robert Godin. Constructing Iceberg Lattices from Frequent Closures Using Generators. 11th International Conference on Discovery Science - DS '08, Oct 2008, Budapest, Hungary. pp.136-147. ⟨inria-00331524⟩