Avoiding the itemset closure computation ”pitfall”

Extracting generic bases of association rules seems to be a promising issue in order to present informative and compact user addedvalue knowledge. However, extracting generic bases requires partially ordering costly computed itemset closures. To avoid the nightmarish itemset closure computation cost, specially for sparse contexts, we introduce an algorithm, called Prince, allowing an astute extraction of generic bases of association rules. The Prince algorithm main originality is that the partial order is maintained between frequent minimal generators and no more between frequent closed itemsets. A structure called minimal generator lattice is then built, from which the derivation of itemset closures and generic association rules becomes straightforward. An intensive experimental evaluation, carried out on benchmarking and ”worst case” datasets, showed that Prince largely outperforms the pioneer algorithms, i.e., Close, A-Close and Titanic.