SMIM Framework to Generalize High-Utility Itemset Mining

In high-utility itemset mining (HUIM), the utility of a set of items is calculated as the sum of the utilities of the individual items. In this paper, we describe scenarios where utility may be less than this sum for multi-item itemsets. To overcome the limitation of the current itemset mining algorithms for such scenarios, we introduce the SMIM framework for itemset mining in which utilities are constrained to be non-negative subadditive and monotone functions over itemsets. SMIM generalizes HUIM, can be used to analyse transaction databases with multi-item discount schemes, and can further be used to mine interesting patterns in a social network dataset. Finally, we explain how to design algorithms for SMIM with any general subadditive monotone utility function.