Pattern Discovery with Utility Occupancy

Jiayi Sun,Wensheng Gan,Jerry Chun-Wei Lin,Han-Chieh Chao

DOI: https://doi.org/10.1109/BigData55660.2022.10020765

IF: 4.426

2022-01-01

Big Data

Abstract:To mine potential and helpful patterns, the majority of studies on pattern discovery from databases have been conducted in the last few decades. They have several obvious drawbacks: 1) Each thing stands out on its own and varies in significance based on factors including utility, risk, interest, and weight. 2) In specific application settings, an object has a favorable or unfavorable effect (e.g., products are often cross-sold and have positive or negative unit profits, which affect benefits). 3) The user could not have all the necessary information because frequent-based patterns typically only include a small percentage of the relevant patterns (for example, occupancy). To address this issue, we apply economic utility theory to the database and data mining fields. We provide a one-phase approach called pnHUO for discovering High Utility Occupancy patterns with positive and negative utility values that beyond frequency and usefulness. According to user interests, frequency, and utility occupancy, there are various utility occupancy patterns with positive and negative utility values. To hold the necessary data, a new frequency-utility tree and an indexed data structure called a positive-and-negative utility-occupancy list are created during the mining process. A number of pruning strategies are further developed using the determined upper bound of utility occupancy to reduce the search space. To evaluate the usefulness and efficiency of the suggested algorithm, five real datasets were tested in experiments, and the results were positive.