資料探勘於旅遊網站顧客關係管理之個案研究

Abstract

由於電子商務的盛行，改變了人們對生活的看法及購物的習慣，並顛覆了現有的經濟及企業經營模式，其中線上旅遊產業是B2C電子商務的最大產業。本研究旨在建立個案旅遊網會員價值模型進而分析會員價值區隔，找出高獲利力會員的特徵並加以分析，同時藉由多變量統計中鑑別分析與類神經網路中倒傳遞網路分析方式加以驗證會員價值模型，綜合以上分析來探討旅遊網顧客關係管理策略以及資料倉儲雛型建立模式建議。本研究針對個案旅遊網資料庫進行資料篩選，採用RFM分析模式，選定消費金額（獲利性）為會員價值模型評量重點，加入RFM模式中會員的消費金額比率的標準差做為權重值，建立會員價值指標模式。 本研究結果與建議如下所述： 一、會員價值指標建立 本研究發現個案旅遊網的76%的收入來自前20%的高價值會員，更顯得高獲利力所佔的20%會員的重要性，顧客關係管理也應從這20%的會員著手。 二、會員消費特徵屬性與高獲利會員價值指標分析 本研究發現個案研究網的高獲利力價值會員的消費屬性特徵如下：性別多為女性、婚姻狀況為未婚、年齡在30~39歲之間、教育程度為大學學歷、職業為服務業居多、月收入在3萬~5萬元之間、大多訂閱旅遊電子報、旅遊線港澳大陸居多、最常購買的旅遊產品為國外航空機票。 三、鑑別分析方法與倒傳遞網路法會員價值模型差異比較 模型驗證結果類神經網路分析模式的整體正確判別率達99.920%，比鑑別分析的96.074%略高。顯示類神經網路的確能更有效進行價值判別。 四、旅遊網資料倉儲建置建議 本研究分析後發現，可以從會員資料庫中篩選出R、F、M三個屬性與會員人口統計變數、會員購買產品類型等加以結合，配合本研究的會員價值指標鑑別模型或會員價值指標類神經網路都可有效的判別出會員的價值，將此資料有規則的長期性紀錄於資料倉儲中，能更有效快速的提供決策者做行銷策略的參考。

關鍵詞：資料探勘、旅遊網站、顧客關係管理、鑑別分析、 RFM模式、倒傳遞網路、會員價值分析Purchasing via the Internet was one of the most rapidly growing forms of shopping, with sale growth rates that outpace buying through traditional retailing, especially in traveling products. In this study, a traveling website’s Customer Value Analysis (CVA) model was proposed and empirically tested by Discriminant Analysis and Back-Propagation Neural Network (BPNN). According to above analysis, we tried to improve travel website customer relationship management strategies and made the suggestion for building data warehouse prototype. The study used the RFM analysis pattern to evaluate the member’s value. We added the standard deviation of the member's expense ratio as the weighted value. The major results were summarized as follows:

1.There are 76% traveling website revenue came from the head 20% valuable members. Most of them were female, unmarried, between 30-39 of the age. Large percentage of these members had university degree, devoted in service industry, earned NT$30,000-NT$50,000 per month, mostly subscribed the travel e-paper. The most popular travel line was HK-Macao-Mainland China. The most often purchased product was foreign airline tickets.

2.Compared neural network model with CVA model, 99.92% of original group cases were correctly classified in neural network model which had higher judgment than discriminant analysis with CVA model, correctly classified as 96.07%.

3.The BPNN can better effectively classified than Discriminant Analysis.

4.This research found that combined the three attributes, R, F, and M, the membership demographic variables, and the member’s purchasing categories with the classified model, it was easily to classify the high potential profitable customer.

Through the classification, it was possible to reach those profitable members. Under the data warehouse, it can much more effectively help the marketing strategies chosen and improve the customer relationship.

Key words: Data Mining, Traveling Website, Customer Relationship Management, Discriminant Analysis, RFM Model, BPN Network, Customer Value Analysis