Scholarlink

<p>Mining frequent high average-utility itemsets (FHAUIs) from quantitative transaction databases is important for utility-oriented analytics because the average-utility criterion highlights utility-efficient purchasing patterns that are often more actionable than conventional high-utility itemsets for applications such as bundle design and promotion planning. However, enumerating all FHAUIs typically produces an overwhelming number of patterns, leading to high runtime and memory consumption and limiting interpretability. This study addresses the efficient discovery of compact representations of FHAUIs, namely closed FHAUIs (CFHAUIs), which eliminate redundancy while preserving support information, and maximal FHAUIs (MFHAUIs), which summarize the largest profitable frequent bundles. Two algorithms are proposed, MC-FHAUIM, which simultaneously mines CFHAUIs and MFHAUIs within a unified process, and C-FHAUIM, which mines CFHAUIs only. Both algorithms follow a prefix-growth search paradigm and are accelerated by five pruning strategies, including two new closure-oriented strategies for early removal of non-closed FHAUI branches. A key technical contribution is a novel weak lower bound (WLB) on the average-utility measure, <i>twlbau</i>, designed to be the tightest WLB; it provides a sound basis for safely pruning branches that may contain FHAUIs but cannot yield closed patterns, while avoiding the risk of missing CFHAUIs under aggressive pruning. Extensive experiments on multiple real and synthetic datasets show that the proposed methods consistently reduce runtime and peak memory usage compared with state-of-the-art algorithms, and exhibit improved scalability as database size increases and thresholds decrease. These results demonstrate that mining CFHAUIs and MFHAUIs is a practical and computationally efficient alternative to enumerating all FHAUIs, enabling concise yet informative outputs for real-world decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient mining of compact high average utility patterns using the tightest weak lower bound

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Thong Tran,
  • Hai Duong,
  • Tin Truong,
  • Bac Le

摘要

Mining frequent high average-utility itemsets (FHAUIs) from quantitative transaction databases is important for utility-oriented analytics because the average-utility criterion highlights utility-efficient purchasing patterns that are often more actionable than conventional high-utility itemsets for applications such as bundle design and promotion planning. However, enumerating all FHAUIs typically produces an overwhelming number of patterns, leading to high runtime and memory consumption and limiting interpretability. This study addresses the efficient discovery of compact representations of FHAUIs, namely closed FHAUIs (CFHAUIs), which eliminate redundancy while preserving support information, and maximal FHAUIs (MFHAUIs), which summarize the largest profitable frequent bundles. Two algorithms are proposed, MC-FHAUIM, which simultaneously mines CFHAUIs and MFHAUIs within a unified process, and C-FHAUIM, which mines CFHAUIs only. Both algorithms follow a prefix-growth search paradigm and are accelerated by five pruning strategies, including two new closure-oriented strategies for early removal of non-closed FHAUI branches. A key technical contribution is a novel weak lower bound (WLB) on the average-utility measure, twlbau, designed to be the tightest WLB; it provides a sound basis for safely pruning branches that may contain FHAUIs but cannot yield closed patterns, while avoiding the risk of missing CFHAUIs under aggressive pruning. Extensive experiments on multiple real and synthetic datasets show that the proposed methods consistently reduce runtime and peak memory usage compared with state-of-the-art algorithms, and exhibit improved scalability as database size increases and thresholds decrease. These results demonstrate that mining CFHAUIs and MFHAUIs is a practical and computationally efficient alternative to enumerating all FHAUIs, enabling concise yet informative outputs for real-world decision-making.