運用資料採礦探討國際電話訊務之流失與挽回

Abstract

我國電信業務自1996年起陸續自由化，在固定網路方面，在2000年釋出3張新的營運執照，得標的民營業者在2001年陸續開始營運。但由於最後一哩鋪設進度緩慢，長途電話利潤較低，而國際電話利潤較高、產品同質性高、消費者轉換提供業者也較為容易，初期新進業者皆採降價方式搶攻客源，使得國際電話業務競爭最為激烈。2004年後，降價空間漸小，市場結構亦漸趨穩定，此階段正是原既有業者應該思考如何開始挽回先前已流失客戶之時機。 故此，本研究擷取原既有業者2004年1月至9月國際電話資料倉儲通話明細資料800餘萬筆，並抽樣出台北縣客戶93餘萬筆資料進行採礦，以羅吉斯迴歸檢定假設，並建構流失客戶的挽回模型。經本研究分析發現，國際電話之價格變化、消費者通訊使用行為和是否能挽回有相關，且價格變化和消費者的通訊使用行為會交互影響通訊量能否挽回。此模型經正確率、反查率、效益值及R平方等指標評估後，確認此模型具備足夠的效用，可提供業者篩選出挽回機率較高之客戶，進行客戶挽回的操作。

The telecom industry in Taiwan has been liberalized from 1996. The government in 2000 issued three licenses of fixed line, and the new carriers have been operated since 1991. Because the Last mile is difficult to construct, the competition of local phone market is not violent. Compared with the long-distance telephone’s less profit, the international telephone’s high profit, homogeneous product properties, and easiness for customers to change the provider shows a powerful attraction. Most new carrier adopted penetration pricing strategy to capture customers and this situation resulted in intense competition between international telephone market. However, while the space of price reduction gets smaller and the market structure tends towards stable after 2004, this is the perfect time for incumbent operator to think how to win lost customers back. We retrieve more than eight millions international call detail records(CDR), which were produced in the period of Jan to Sep in 2004, from incumbent operator’s Data warehouse. After sampling 930,000 records that coutomer born in Taipei,we use Data mining methodology to analyze it. After that, we use Logistic regression to test hypotheses and construct regression model. Finally, it is found that the variation of international call price and communication useage behavior is related to whether the lost customers can be recaptured or not. Additionally, both factors have interactive correlation to affect customer’s choice. Response rate, Recall rate, Lift, and R square verify this model’s usefulness, and we suggest incumbent operator use this model to recapture lost traffic.