Adaptive constraint pushing in frequent pattern mining

Abstract

Pushing monotone constraints in frequent pattern mining can help pruning the search space, but at the same time it can also reduce the effectiveness of anti-monotone pruning. There is a clear tradeoff. Is it better to exploit more monotone pruning at the cost of less anti-monotone pruning, or viceversa? The answer depends on characteristics of the dataset and the selectivity of constraints. In this paper, we deeply characterize this trade-off and its related computational problem. As a result of this characterization, we introduce an adaptive strategy, named ACP (Adaptive Constraint Pushing) which exploits any conjunction of monotone and anti-monotone constraints to prune the search space, and level by level adapts the pruning to the input dataset and constraints, in order to maximize efficiency.