Pushing Tougher Constraints in Frequent Pattern Mining

In this paper we extend the state-of-art of the constraints that can be pushed in a frequent pattern computation. We introduce a new class of tough constraints, namely Loose Anti-monotone constraints, and we deeply characterize them by showing that they are a superclass of convertible anti-monotone constraints (e.g. constraints on average or median) and that they model tougher constraints (e.g. variance or standard deviation) which have never been studied before. Then we show how these constraints can be exploited in a levelwise Apriori-like computation by means of new datareduction strategies, outperforming previous algorithms for tough constraints, and exploiting much tougher ones with the same effectiveness.