A Global Constraint for Closed Frequent Pattern Mining

Abstract

Discovering the set of closed frequent patterns is one of the fundamental problems in Data Mining. Recent Constraint Programming (CP) approaches for declarative itemset mining have proven their usefulness and flexibility. But the wide use of reified constraints in current CP approaches leads to difficulties in coping with high dimensional datasets. In this paper, we proposes the ClosedPattern global constraint to capture the closed frequent pattern mining problem without requiring reified constraints or extra variables. We present an algorithm to enforce domain consistency on ClosedPattern in polynomial time. The computational properties of this algorithm are analyzed and its practical effectiveness is experimentally evaluated.

CoverSize: A Global Constraint for Frequency-Based Itemset Mining

“…Earlier papers on using CP for itemset mining focus mostly on generality and decompose the itemset mining constraints into many (reified) linear constraints [19] at the cost of efficiency. In line with recent works in CP for sequence mining [3,4,22], Lazaar et al [24] have shown that a single global constraint for closed frequent itemset mining can outperform a decomposition approach. This comes at significant cost for generality though, because 1) by encapsulating all but the itemset variables, only syntactic constraints on the items can be added; 2) only closed frequent patterns can be found and adding syntactic constraints can have unwanted side-effects [6].…”

“…Guns et al [19] showed that FIM problems could be modelled and solved using Constraint Programming (CP) with the additional benefit that new constraints can easily be integrated into the models. Since then several CP (also SAT) approaches have been proposed for other data-mining problems such as frequent sequence mining [22,29], dominance-based pattern mining [28] and closed FIM [20,21,24].…”

Constraint Programming for Multi-criteria Conceptual Clustering

“…We also plan to evaluate scale-up properties of ILP for this problem, and combine ILP with CP if we observe complementary performance. Finally, we plan to evaluate the interest of combining our CP model with the propagation algorithm of [14].…”

“…These constraints are used to filter the search space during the mining process, and allow CP to be competitive with dedicated mining tools such as LCM. Most recently, [14] introduced a global constraint for extracting frequent closed itemsets. This global constraint enforces domain consistency in polynomial time, and it is quite competitive with LCM: if it is an order slower on basic queries, it is more efficient for complex queries where extra constraints are added.…”

User’s Constraints in Itemset Mining

“…In a recent line of work [7,4,5,3,8], constraint programming (CP) has been used as a declarative way to solve data mining problems. Such an approach has not competed yet with state of the art data mining algorithms [11,9] for simple queries.…”