Itemset Mining with Penalties
Abstract
We introduce a preferences-based itemset mining framework. Preferences are encoded by a penalty function over the transactions in a database. We define an itemset mining problem where we associate to each transaction a penalty value. This problem consists in generating the frequent itemsets with a maximum penalty threshold. We then provide a propositional satisfiability based encoding. We extend the previous problem with a penalty function over items, where we use two maximum penalty thresholds, over the transactions and over the items. In this setting, computing the optimum itemsets corresponds to computing Pareto front. The experimental evaluation on real world data shows the feasibility of our approach.