On maximal frequent itemsets mining with constraints

Recently, a new declarative mining framework based on constraint programming (CP) and propositional satisfiability (SAT) has been designed to deal with several pattern mining tasks. The itemset mining problem has been modeled using constraints whose models correspond to the patterns to be mined. In this paper, we propose a new propositional satisfiability based approach for mining maximal frequent itemsets that extends the one proposed in [20]. We show that instead of adding constraints to the initial SAT based itemset mining encoding, the maximal itemsets can be obtained by performing clause learning during search. A major strength of our approach rises in the compactness of the proposed encoding and the efficiency of the SAT-based maximal itemsets enumeration derived using blocked clauses. Experimental results on several datasets, show the feasibility and the efficiency of our approach.