線上消費者平台再購行為的RFM預測模型－以Yahoo！奇摩拍賣女裝為例

A RFM-Based Forecast Model of Consumers' E-marketplace Repurchase Behavior at Yahoo! Taiwan Auction Website

何靖遠(C. Y. Ho) ； 陳慧玲(H. L. Chen) ； 廖致淵(C.Y. Liao)

《Journal of Data Analysis》 9卷1期 (2014/02) Pp. 1-23

https://doi.org/10.6338/JDA.201402_9(1).0001

摘要

1990年哈佛商業評論的一篇文章揭露跨產業的研究顯示，增加5%的顧客留住率可以提高獲利25%至95%。雖然資策會預估台灣2011年的電子商務產值可達4,300億元的規模，實際卻觀察到2009年只有28.6%的店家獲利，網路上的平台業者和賣家要獲利，如何開發新顧客並留住舊顧客變成非常重要的課題。本研究將線上消費者在同一購物平台重複消費的現象定義為平台的再購，表示線上消費者對該平台的一種行為忠誠。我們以資料庫行銷的概念RFM為基礎，使用交易間隔天數（R）、累積交易次數（F）、累積交易金額和平均交易金額（M）等四個RFM變數，加上消費者在平台的交易賣家數和最近一次給予的評價等共六個變項，預測線上消費者未來的平台再購行為。本研究以網頁內容探勘技術耙取Yahoo！奇摩拍賣購物平台的女裝上衣資料，經隨機選取部分賣家，取得其五個月期間的18,881筆交易資料，找出5,682個有購買且給予評價的買家。本研究以SPSS的邏輯斯迴歸和STATISTICA統計軟體的隨機森林方法分析，結果顯示六個預測變項對平台的再購行為均有顯著影響，唯平均交易金額的影響是負向的，與累積交易金額不同。對照針對賣家再購行為預測模型的研究結果，在預測模型中的四個RFM預測變數中，除累積交易次數以外，其餘預測變數對平台再購的影響程度均比對賣家再購的影響為小，而買家最近一次所給予評價的影響在平台再購行為預測模型中的影響程度也較小。由於在電子商務再購行為這方面的文獻是比較少的，本研究對於使用實際交易資料的學術研究有一定的貢獻。實務上本研究結果有助於網路購物之平台業者和賣家掌握關鍵資訊以有效留住舊顧客和開發新顧客。

關鍵字

平台再購行為 ； 預測模型 ； 隨機森林

並列摘要

Revealed by a Harvard Business Review article in 1990, ”increasing customer retention rates by 5% increases profits by 25% to 95%” is shown by a cross-industry study. The MIC of III in Taiwan estimated the e-commerce trading volumes can reach 430 billion NT dollars in 2011, yet only 28.6% online sellers can make a profit in 2009. How to attract new customers and retain existing ones becomes a critical issue for e-marketplace operators and sellers as well. This study defines the repeated shopping of online consumers in a certain website as his/her e-marketplace repurchase behavior, which reflects the behavioral loyalty to e-marketplace. Based on the database marketing concept of RFM, we establish a forecast model for consumers' e-marketplace repurchase behavior with six predictor variables associated with an online consumer. They include the number of days since last purchase (recency), cumulative number of purchases (frequency), cumulative and average purchasing amount (monetary), number of repurchased sellers, and the latest rating given in the website.The study used web content mining to crawl transactions of women's clothing at Yahoo! Taiwan Auction website. We captured 18,881 transactions in a five-month period from randomly selected sellers of women's clothing and identified 5,682 buyers who purchased from those sellers and posted ratings. The Logistic regression in SPSS and the Random Forests in STATISTICA were used for data analysis, and the results showed all six predictor variables significantly influenced e-marketplace repurchase behavior, only the effect of average purchasing amount is negative, which is different from that of cumulative purchasing amount. In contrast to the findings by the forecast model of seller repurchase behavior, the impacts of the four RFM variables except the cumulative number of purchases are smaller in e-marketplace repurchase behavior than in the seller's repurchase behavior and the impact of latest rating is also smaller in the forecast model of e-marketplace repurchase behavior. Due to the literature of repurchase behavior in e-commerce is scant, this study further the research streams in using actual trading data. In practice, it can facilitate e-marketplace operator and sellers focusing on the critical information to retain existing customers and attract new ones.

並列關鍵字

E-marketplace Repurchase Behavior ； RFM ； Forecast Model ； Random Forests

參考文獻

何靖遠、廖致淵、許文錦()。

Anderson, R. E.,Srinivasan, S. S.(2003).E‐satisfaction and e‐loyalty: A contingency framework.Psychology and Marketing.20(2),123-138.

Baucer, C.L.(1988).A Direct mail Customer Purchase Model.Journal of Direct Marketing.2(3),16-24.

Bonabeau, E.(2004).The perils of the imitation age.Harvard business review.82(6),45-54.

Breiman, L.(2001).Random Forest.Machine learning.45,5-32.