A framework for redescription set construction

Redescription mining is a field of knowledge discovery that aims at finding different descriptions of similar subsets of instances in the data. These instances are characterized with descriptive attributes from one or more disjoint sets of attributes called views. By exploring different characterizations it is possible to find non trivial and interesting connections between different subsets of attributes. In this work, we explore the process of creating possibly large and heterogeneous redescription set in which redescriptions are iteratively improved by a conjunctive refinement procedure aimed at increasing redescription accuracy. This set is used by our redescription set construction procedure to create multiple redescription sets of user defined size. Set construction is based on redescription selection by using multi-objective optimization incorporating user defined importance levels towards one or more redescription quality criteria. These properties distinguish our approach from current state of the art approaches that create one, mostly smaller set that contains redescriptions satisfying a pre-defined set of constraints. We introduce a new redescription quality criterion that assesses the variability of redescription accuracy when missing values are present in the data. Finally, we compare the performance of our framework with three state of the art redescription mining algorithms.