Discovering Top-k Profitable Patterns for Smart Manufacturing
Shicheng Wan, Jiahui Chen, Peifeng Zhang, Wensheng Gan, Tianlong Gu · Companion Proceedings of the Web Conference 2022 · 2022
In the past, many studies were developed to discover useful knowledge from rich data for decision-making in wide applications in Internet of Things and Web of Things, such as smart manufacturing. Utility-driven pattern mining (UPM) technology is famous in knowledge discovering domain. However, one of the biggest issues of UPM is the setting of a suitable minimum utility threshold (minUtil). The higher minUtil is, the fewer interesting patterns are obtained. Conversely, the lower minUtil is, the more useless patterns are discovered. In this paper, we propose a solution for discovering top-k profitable patterns with average-utility measure, which can be applied to manufacturing. The average-utility of a pattern, w.r.t its corresponding length, can be used to fairly measure the pattern. The proposed new upper-bounds on average-utility are tighter than previous upper-bounds. Moreover, based on these upper-bounds, the novel PPT algorithm utilizes merging and projection techniques to greatly reduce the search space. By adopting several threshold raising strategies, the PPT algorithm can discover correct top-k patterns in a short time. We also implemented the efficiency and effectiveness of the algorithm on real and synthetic datasets. The experimental results reveal that the algorithm not only get a complete set of top-k interesting patterns, but also works better than the state-of-the-art algorithm in terms of runtime, memory consumption and scalability. Especially, the proposed algorithm performs very well on dense datasets.