運用資料採礦技術於零售業之研究—以居家用品專賣店為例

Applying Data Mining Technology to Retail Industry–A Case Study on Home Furnishing Stores

吳岱壎(Tai-Hsun Wu)

指導教授 ： 吳泰熙

國立臺北大學/商學院/企業管理學系/碩士(2012年)

摘要

隨著電腦硬體設備的發展，資料的儲存容量快速成長，企業組織不僅要能有效儲存、管理資料，還需要在海量資料中尋找出有價值的資訊，並幫助做出下一步行動的決策。零售業直接接觸終端消費者，如果能透過資料採礦技術在大量的顧客資料與交易紀錄中挖掘出特徵、規則或趨勢，將可提供更具說服力的行銷策略之建議。
本研究以某居家用品專賣店為例，由其聯名卡顧客資料與消費紀錄，採用標準化之RFM消費行為變數，並透過AHP層級分析法評估RFM之權重，以建立符合零售業特質之顧客價值量化模型。透過顧客價值量化模型之建立，以之為顧客分級的基礎。顧客分級係為幫助企業掌握重要的忠實顧客群，將資源做最有效的配置，並得到更高的回應率。本研究以K-Means法進行顧客集群分析，以標準化之RFM變數將個案公司會員分為「忠實顧客」、「目標顧客」、「流失顧客」三大族群。並由顧客價值量化模型中取相同比例人數進行交叉分析，發現兩種分群方式雷同率達92%。
本研究根據零售業的產業特性，透過資料採礦之關聯規則與時序分析，發掘適合顧客群的商品組合，找出哪些商品最常一起被購買，哪些商品存在購買的順序關係，並將結果運用於交叉銷售與垂直銷售規劃。零售業銷售品項數量龐大，要尋找出有效的關聯規則有賴於合適的分類架構。本研究依個案公司提供之中分類、小分類、產品別進行關聯規則探勘外，另增加細分類關聯規則研究。對關聯規則的選取，也以支援、機率與重要性等指標分別檢視，可提供個案公司進行產品組合與行銷活動規劃之參考。

關鍵字

資料採礦 ； RFM ； AHP ； 集群分析 ； 關聯規則 ； 時序分析

並列摘要

With the development of computer hardware, data storage capacity and capability are fast-growing; organizations not only be able to effectively store, manage data, it also need to find out valuable information in the big data, and to help assist in making the next critical decision. Retail industries usually have contact with the end users, through data mining techniques, will be able to gather a lot of customer’s information, transaction records, buying patterns, rules, or trends, and these information will provide more convincing marketing strategy recommendations.
In this study, a home furnishing store, for example, through its joint co-branded cards, it will provide customer’s information and transaction records, using standardized the RFM consumer behavior variables and weights of the method of assessment of RFM, to create customer value in line with the retail characteristics of quantitative analysis through the AHP model. Through the establishment of quantitative models of customer value to the customer cluster on the basis of customer grading system to help businesses master a loyal customer base, the resources to do the most effective configuration, and a higher response rate. In the study, the K-Means method, the customer cluster analysis standardized the RFM variables to the case company is divided into three major groups; "loyal customers", "target customers", "lost customers," By the quantitative model of customer value to take the same proportion the number of cross-analysis, we found that up to 92% similarity with the two methods.
In this study, based on the retail industry characteristics, through data mining of association rules and sequential patterns analysis, to explore suitable product mix for the customer, find out which product mix most frequently purchased together, which goods exist to purchase the order of relations, and apply the results of cross-selling and up-selling planning. Due to diverse numbers of product items in retail sales, to effectively find out the valuable association rules greatly depends on the appropriate category structure. In this study, according to the case company being classified, classification by product association rule mining, and to increase the fine classification association rules. In the selection of association rules, probability, and importance, the indicators for analysis, to provide reference to the product mix and marketing activities planning.

並列關鍵字

Data Mining ； RFM ； AHP ； Cluster Analysis ； Association Rules ； Sequential Patterns Analysis

參考文獻

林祥生、劉益豪(2008)。應用資料採礦探討國際線航空旅客之線上購票行為。運輸計劃季刊，37(2)，197-236。

張心馨、蔡獻富(2004)，以Data Mining技術結合SOM和K-Mean的消費者分群方法於顧客關係管理和績效獲利性評估之實證研究。資訊管理學報， 11(4)，161-203。

許雅涵、徐武永與李世能( 2011)。運用RFM模型分析顧客消費行為與貢獻度之研究及應用實務－以中油會員卡顧客為例。石油季刊，47(1)， 83-100。

郭瑞祥、蔣明晃與陳宏毅(2003)。顧客價值分析之隨機模型建立及實證。管理學報，21(5)，675-692。

劉蕙(2006)。以群集分析與序列型樣輔助資料庫行銷策略。元智大學資訊管理學系碩士論文。