Conclusions
In this paper, we proposed an algorithm for efficiently mining weighted erasable patterns over sliding window-based product data streams. In order to allow the proposed algorithm to effectively consider the latest information on a given product data stream, we devised new tree and list data structures and effectively applied them into our mining process. Through the window updating and tree restructuring techniques, we also allowed the proposed tree structure to efficiently maintain the latest information in a compact tree form. Moreover, we improved the mining performance of the proposed algorithm by storing only essential information of the tree’s data used in the actual mining process into the list structure and employing this information during the pattern expanding operations. In addition, our algorithm mined more meaningful patterns considering not only the gain values of patterns but also different item importance by applying the weight factors into the erasable pattern mining framework. We also enhanced the mining performance of our algorithm by proposing a strong pattern pruning technique using the weight factor, and prevented unintended pattern losses caused by the weight factor through applying the overestimated method satisfying the anti-monotone property. The extensive performance evaluation results clearly supported the contributions of this paper. From the results and analyses of the performance evaluations, we determined that the proposed method is more efficient and scalable than the previous state-of-the-art approaches in terms of runtime, memory, and pattern generation aspects.
บทสรุปIn this paper, we proposed an algorithm for efficiently mining weighted erasable patterns over sliding window-based product data streams. In order to allow the proposed algorithm to effectively consider the latest information on a given product data stream, we devised new tree and list data structures and effectively applied them into our mining process. Through the window updating and tree restructuring techniques, we also allowed the proposed tree structure to efficiently maintain the latest information in a compact tree form. Moreover, we improved the mining performance of the proposed algorithm by storing only essential information of the tree’s data used in the actual mining process into the list structure and employing this information during the pattern expanding operations. In addition, our algorithm mined more meaningful patterns considering not only the gain values of patterns but also different item importance by applying the weight factors into the erasable pattern mining framework. We also enhanced the mining performance of our algorithm by proposing a strong pattern pruning technique using the weight factor, and prevented unintended pattern losses caused by the weight factor through applying the overestimated method satisfying the anti-monotone property. The extensive performance evaluation results clearly supported the contributions of this paper. From the results and analyses of the performance evaluations, we determined that the proposed method is more efficient and scalable than the previous state-of-the-art approaches in terms of runtime, memory, and pattern generation aspects.
การแปล กรุณารอสักครู่..
