Title:Towards Sequence Utility Maximization under Utility Occupancy Measure

Abstract:The discovery of utility-driven patterns is a useful and difficult research topic. It can extract significant and interesting information from specific and varied databases, increasing the value of the services provided. In practice, the measure of utility is often used to demonstrate the importance, profit, or risk of an object or a pattern. In the database, although utility is a flexible criterion for each pattern, it is a more absolute criterion due to the neglect of utility sharing. This leads to the derived patterns only exploring partial and local knowledge from a database. Utility occupancy is a recently proposed model that considers the problem of mining with high utility but low occupancy. However, existing studies are concentrated on itemsets that do not reveal the temporal relationship of object occurrences. Therefore, this paper towards sequence utility maximization. We first define utility occupancy on sequence data and raise the problem of High Utility-Occupancy Sequential Pattern Mining (HUOSPM). Three dimensions, including frequency, utility, and occupancy, are comprehensively evaluated in HUOSPM. An algorithm called Sequence Utility Maximization with Utility occupancy measure (SUMU) is proposed. Furthermore, two data structures for storing related information about a pattern, Utility-Occupancy-List-Chain (UOL-Chain) and Utility-Occupancy-Table (UO-Table) with six associated upper bounds, are designed to improve efficiency. Empirical experiments are carried out to evaluate the novel algorithm's efficiency and effectiveness. The influence of different upper bounds and pruning strategies is analyzed and discussed. The comprehensive results suggest that the work of our algorithm is intelligent and effective.