A parallel declarative framework for mining high utility itemsets
Résumé
One of the most active research topics in data mining is pattern discovery involving the well-known task of enumerating interesting patterns from databases. The problem of mining high utility itemsets is to find the set of items with the highest utility values based on a given minimum utility threshold. However, due to the advancement of big data technologies, finding all itemsets is much more harder due to the huge number of patterns and the large required resources. Parallel processing is an effective way to efficiently address the problem of mining patterns from large databases. Based on classical propositional logic, we propose in this paper a parallel method to handle efficiently the problem of discovering high utility itemsets from transaction databases. To do this, a decomposition technique is used to splitting the original problem of mining high utility itemsets into smaller and independent sub-problems that can be handled easily in a parallel manner. Then, empirical evaluations on different real-world datasets show that the proposed method is very efficient while being flexible enough to handle additional user constraints when discovering closed high utility itemsets.