探勘含重點項目集之關聯法則─以賣場資料為例

Mining Association Rules with Focus Itemset─Using Basket Data

林錦誼(Gin-Yi Lin)

指導教授 ： 林志麟

元智大學/資訊學院/資訊管理學系/碩士(2001年)

摘要

關聯法則為目前資料探勘領域相當熱門之研究方向。然而，多數研究皆著眼於以單一最小支持度進行資料篩選的根基上，卻忽略此方式可能導致稀少項目問題（rare item problem）及使用者偏好問題（users' interests problem）。以前者而言，某些項目（item）可能天生支持度較低，因此以單一最小支持度進行篩選時較易被剔除，如：高價位商品、新產品。以後者而言，使用者可能有自身之偏好，但傳統關聯法則卻無法兼顧此方面的考量。
本研究提出一改進之關聯法則演算法，稱為聚焦式關聯法則（Focus Association Rules）。此演算法可彌補上述關聯法則不足之處。簡而言之，聚焦式關聯法則採用多重最小支持度法（multiple minimum supports），即每一項目有其適用之最小支持度。因此當進行資料篩選時，不同項目集（itemset）將可以不同標準衡量。所有項目集之最小支持度皆由聚焦式關聯法則依據使用者設定之單一最小支持度自動產生。對於使用者而言，操作方式和傳統關聯法則幾乎相同。故相較於其它類似研究，聚焦式關聯法則擁有較高的可行性，且對於使用者而言進入門檻較低。

關鍵字

資料探勘 ； 關聯法則 ； 聚焦式關聯法則 ； 稀少項目問題 ； 使用者偏好問題 ； 多重最小支持度

並列摘要

Association rules are one of the most popular research areas in data mining. However, most researches are based on a uniform minimum support and ignore the rare item and users' interests problems it brings. For the former, some items may have lower support naturally and would be filtered out easily when using a uniform minimum support, such as the high-priced, new items, etc. For the latter, users may have their special interests, but traditional association rules can not take those factors into account.
This research proposes a new algorithm, namely the Focus Association Rules to deal with these problems. In brief, the Focus Association Rules adopt the concept of multiple minimum supports, i.e. each item has its own minimum support, and judge itemsets according to their own criterion. The minimum support of each itemset is generated by the Focus Association Rules automatically. Users only need to specify one minimum support as before. Therefore, comparing with other related literatures, the Focus Association Rules are more practical and easier to users.

並列關鍵字

data mining ； association rules ； focus association rules ； rare item problem ； users' interests problem ； multiple minimum supports

參考文獻

[3]Database Research Group, Department of Computer Science and Engineering, The Chinese University of Hone Kong, http://www.cse.cuhk.edu.hk/~kdd/.

[5]H. Mannila, "Database Methods for Data Mining", KDD-98 Tutorial, 1998. page

[7]Jiawei Han and Micheline Kamber, Data Mining: Concepts and Techniques, Morgan Kaufmann Publishers, 2000.

[9]Ke Wang, Yu He and Jiawei Han, "Mining Frequent Itemsets Using Support Constraints", the Proceedings of the 26th VLDB Conference, Cairo, Egypt, pp.43-52, 2000.

[19]Show-Jane Yen and Arbee L.P. Chen, "An Efficient Data Mining Technique for Discovering Interesting Association Rules", the Proceedings of the 8th International Workshop on Database and Expert Systems Application, pp.664-669, 1997.