應用雙變量層級貝氏定理於顧客價值分析─以網路購物為例

Abstract

企業間的競爭日趨激烈，顧客的需求也越來越多樣化，企業為了能夠吸引顧客、維繫顧客關係，紛紛投入大量行銷資源於顧客身上，然而如何區辨金質顧客使得資源配置的效率極大化，便成為現今企業在進行顧客關係管理所關心的議題。本研究針對此問題，運用層級貝氏模式 (Hierarchical Bayesian Methodology) 推導顧客購買行為預測模型，並以實證資料佐證本研究模型之適用性。此模型能對每位顧客進行個別預測，因此能達成以一對一行銷為基礎的顧客關係管理。 本研究方法依以下步驟進行： 1. 針對個別顧客進行「每次購買金額」與「購買期間」分析 2. 建構顧客購買行為分析之雙變量層級貝氏模型 3. 運用層級貝氏理論來進行顧客個人參數分配之推導 4. 運用馬可夫鏈蒙地卡羅法進行顧客個人參數之估計 5. 預測顧客個人購買行為 基於上述之方法，利用某購物網站之實際交易紀錄進行本模型之資料驗證，結果顯示本研究所建立之顧客購買行為模型對於顧客購買行為的描述更為準確，並能捕捉顧客異質性，在動態資料庫配合下，將可即時更新模型進行預測，更能有效預測個別顧客的購買行為。

The competition among enterprises is becoming fiercer while demands of customers also become more and more diversified. In order to attract customers and maintain the relationship with them, enterprises must put a lot of marketing resources in customers. Therefore, how to distinguish the most valuable customer to maximize the efficiency of allocating resources is becoming one of the most important topics that enterprises are interested in now. This thesis uses Hierarchical Bayesian methodology to derive the model for individual customer value estimation. This model is also validated by on-line shopping data. Therefore the model can be used to facilitation one-to-one customer relationship management. This proposed methodology consists of five steps: 1. Analyze customer’s purchase amount and inter-purchase time. 2. Model each customer’s purchase behavior using the Bi-variate Hierarchical Bayesian methodology. 3. Derive the posterior distributions for each purchase model. 4. Use Markov Chain Monte-Carlo methodology to estimate the distributions’ parameters. 5. Predict each customer’s value by estimating their purchasing amounts. The results of this validation study confirms that our model can capture customer behavior more precisely than other models. Besides, this model can also catch heterogeneity between customers.