The Application of Apriori on Customized Shopping Behavior

[[abstract]]資料探勘是致力於資料分析和理解、揭示資料內部蘊藏知識的技術，經過多年努力，資料探勘發展出許多新概念和方法。特別是近年來，一些基本概念和方法趨於清晰，正朝向更深入的方向發展。從目前現狀來看，關聯規則是資料探勘中成果頗豐且較活躍的研究分支，留給研究者許多可更深入的課題。面對個人真實購物行為的原始資料庫，關聯規則在其中所扮演的角色及閾值設限所可能造成的分析結果仍不明確。因此，鑒於當前資料探勘技術和關聯規則研究的現狀和發展趨勢，本文選擇該課題開展相關研究工作。 本文的研究工作首先介紹資料探勘技術與關聯規則中Apriori演算模型的發展歷史與其相關應用。其次，本文設計出可適用於真實世界中的資料庫模板，針對個人在賣場內的真實購物行為進行詳盡記錄。然後，本文給出廣義關聯規則信賴度閾值設限的構想，在幾乎不對關聯規則信賴度閾值設限的情況下，取得有購買之商品組合的全部關聯規則進行預測正確度分析工作。此外，本文根據頻繁項集所產生的關聯規則信賴度走勢對商品組合購買變化進行深刻探討。結果表明，關聯規則的信賴度應設閾值將隨不同研究對象在不同購物環境下產生明顯差異，且信賴度走勢須配套觀察才可得知真正的商品組合購買結果。本文在分析資料探勘技術之Apriori演算模型系統的基礎上，導入個人於賣場內的真實購物原始資料。在不對信賴度閾值設限的條件下，進行關聯規則預測正確度分析及信賴度走勢分析兩方面的研究，所得到的結論在可用性方面具有極佳的應用前景。[[abstract]]Data systems nowadays have been widely adopted in businesses. The massive data collected from business transactions have made entrepreneurs realize to use the data for supporting their business decision makings. Knowledge Discovery in Database (KDD) therefore becomes a wave of essential concept to extract useful and valid knowledge from data even though the knowledge is intuitive or previously unknown. The processes and algorithms of data mining aim to find patterns that describe underlying relationships in the data, for example, association rules (Agrawal, 1993) and clustering. Association rule mining looks for frequently occurring patterns in the data and is often used for market basket analysis. The famous result is the diaper-beer rule in the basket based on association rule. Benefits for retailers are better shelf managing, goods supplying, and market promotion. Customization techniques, to accommodate differences between individuals, enable the dynamic insertion, personalization, or suggestion of contents in any format that is relevant to the individual user, based on the user’s implicit/explicit behavior and preferences. While customers are claimed to be beneficial from data mining, the purpose of this research is to forecast individual customer’s shopping behavior by adopting Apriori algorithm. In addition to suggest how to decide an interestingness index for producing association rules for individual customers, this study also provides various principles to interpret changes on each customer’s shopping goods in her/his shopping basket. Providing a reliable shopping list suggestion for individual customer before her/his entering the store, designing a better customized promotion for individual customer, and attracting customers visiting the store repeatedly would be on the corner for the retailers