To Örebro University

Pattern mining provides useful tools for exploratorydata analysis. Numerous efficient algorithms exist that are ableto discover various types of patterns in large datasets. However,the problem of identifying patterns that are genuinely interestingto a particular user remains challenging. Current approachesgenerally require considerable data mining expertise or effortand hence cannot be used by typical domain experts.

We show that it is possible to resolve this issue by interactivelearning of user-specific pattern ranking functions, where a userranks small sets of patterns and a general ranking function isinferred from this feedback bypreference learningtechniques.We present a general framework for learning pattern rankingfunctions and propose a number of active learning heuristicsthat aim at minimizing the required user effort. In particular wefocus on Subgroup Discovery, a specific pattern mining task.

We evaluate the capacity of the algorithm to learn a ranking ofa subgroup set defined by a complex quality measure, given onlyreasonably small sample rankings. Experiments demonstrate thatpreference learning has the capacity to learn accurate rankingsand that active learning heuristics help reduce the requireduser effort. Moreover, using learned ranking functions as searchheuristics allows discovering subgroups of substantially higherquality than those in the given set. This shows that activepreference learning is potentially an important building blockof interactive pattern mining systems.