周世玉;蕭登泰;
頁: 183-199
日期: 2005/04
摘要: 顧客交易資料庫蘊含對實施顧客關係管理不可或缺的重要資訊。本研究利用國內某網路電話公司顧客交易資庫的非契約型顧客實際交易資料，估計個別顧客關係危險率，進而分析顧客關係長度與顧客終身價值，以及建構各類顧客之顧客關係危險率預測模式。研究結果指出，利用倒傳遞網路估計個別顧客關係危險率時，訓練集合的MSE與錯誤率分別為0.000077與1.30631%，而測試集合的MSE與錯誤率則分別為0.002469與3.636028%，皆顯示出不錯的績效。估計顧客關係長度中位數的平均與剩餘關係長度中位數的平均分別為9.14個月與1.97個月。在顧客終身價值方面，平均每位顧客的價值約為964元，共可產生約615萬元的價值。以各期估計自我相關係數與偏自我相關係數為自變數對顧客進行分群，將顧客分為兩群，集群一的顧客顧客危險率預測適合AR(1)模式，而集群二的顧客顧客危險率預測則較適合白噪音模式。
關鍵字: 顧客關係管理;資料庫;類神經網路;

A Data Mining Study on Customer Transaction Data Base-Using Non-contracted Customers of a Domestic I-Phone Company as an Example

Abstract: Customer transaction databases contain relevant information that is related to the implementation of customer relationship management practice. In this study, we use transaction records of non-contract type customers of a domestic internet phone company to estimate individual customer relationship hazard rates, analyze customer lifetime length and lifetime value, and construct customer relationship hazard forecast models for different types of customers. The empirical results indicate that the MSE and error rate are 0.000077 and 1.30631% for the training set, and 0.002469 and 3.636028% for the testing data set by using the back propagation neural network technique to estimate customer relationship hazard rates. Both highlight good performance. The medians of average customer relationship length and average remaining relationship length are 9.14 and 1.97 months respectively. The average customer value is about 964 dollars which is equivalent to 615 million dollars in total. Customers are separated into two groups by using autocorrelation functions and partial autocorrelation functions. The customer relationship hazard rate forecasting models for the two groups are AR(1) and white noise models, respectively.
Keywords: Customer Relationship Management;Database;Neural Network;

瀏覽次數: 10045     下載次數: 73

引用     導入Endnote