| 摘要: | 近年來電子商務環境的營業額成長快速，預估在2020年虛擬通路營業額更會成長至整個零售業的20%，因此企業應重視虛擬通路的經營。在台灣電子商務環境中，買家最常進行交易的平台有Yahoo!奇摩、露天以及PChome。由於購物平台之間的競爭相當激烈，我們透過觀察購物平台上消費者的再購行為，希望幫助業者找出影響買家再購行為的因素，設法留住這些有價值的顧客。根據哈佛商業評論一份跨產業調查報告，如果買家的再購率上升5%，整體獲利則可以上升25%至95%，再購率對獲利的正向影響在電子商務將更為可觀。因此本研究將買家在同一購物平台重複交易的行為視為平台再購行為，有平台再購行為的買家視為有價值的顧客。
本研究蒐集Yahoo!奇摩拍賣16個類別的商品，自2014年1月1日至2015年4月30日的所有交易資料，約1,800萬筆。本研究延續過去對平台再購的研究，以2015年1月1日為預測時間點，之前3個月的交易當做歷史資料，之後2個月的交易作為未來資料，產生9個預測變數，包括RFM相關的四個變數以及擴充的五個變數和平台再購變數。我們先進行敘述統計分析，進而分別針對女裝與服飾類別、四類熱門經驗品類別以及全經驗品類別共八類商品建立平台再購行為的預測模型，以探討此模型在多重類別商品中的準確率。我們發現在全經驗品再購預測分析中，若交易間隔天數越長、平均交易金額越高及商品總數量越多，未來再購機率越低；另一方面，當消費者在過去一段時間中購買的次數越多、總交易金額越高、最後一次評價越高、交易賣家數越多、商品類別多樣性越高，未來再購機率越高。在預測方法選擇上，決策樹較羅吉斯迴歸有更準確的預測結果。基於本研究的結果，我們提供平台更準確找出有價值買家的方法，並給予實務上的建議。;E-commerce revenue has grown rapidly in recent years. It is estimated that revenue from the virtual channel in retailing industry will reach twenty percent in 2020. Enterprises should pay more attention to the operation of the virtual channel. Buyers in Taiwan are most commonly making online transactions on Yahoo!, Ruten and PChome. Given there is a fierce competition between e-commerce platforms in Taiwan, we hope to understand factors influencing buyers’ repurchase behaviors so to help identify valuable online customers. Based on a cross-industrial study in Harvard Business Review, if buyers’ repurchase rate increase five percent, the overall revenue can increase twenty five percent to ninety five percent, and this positive effect would be greater for e-businesses. Therefore this study defines repeated purchase on a certain e-commerce website as the platform repurchase behavior of online consumers, and who are considered valuable customers to the e-commerce website.
This study collected all transaction data from Yahoo! Taiwan auction website, including sixteen categories of products from January 1, 2014 to April 30, 2015, totaled 18 million transaction records. As a sequel to previous studies in platform repurchase behavior, we use January 1, 2015 as the prediction point in time, and three preceding months of transaction records as historical dataset and two succeeding months of transaction records as future dataset to generate 9 independent variables, including 4 RFM-related variables and 5 augmented variables, and the repurchase behavior variable. We conduct descriptive statistical analysis on 16 categories of data and then establish the prediction models of platform repurchase behavior in women apparel category, in popular experience goods categories and in entire experience goods categories respectively, to examine the effectiveness of the prediction model across multiple product category. Our findings show that recency, average monetary and purchase quantity are negatively related to repurchase behavior, while frequency, total monetary, last rating, number of purchased sellers, category diversity are positively related to repurchase behavior. In terms of analytical methodologies, we find that decision tree performs better than logistic regression. Based on the study results, we provide guidelines to identify valuable buyers and practical advices to the platform. |
