近年來電信市場蓬勃發展，各家業者致力於各種行銷方式，目的是為了吸引更多客戶加入。在眾多的行銷企畫中，雖能夠吸引不同的消費者，但也會耗費業者高成本資源。倘若能以較低的成本達到相同的結果，對業者而言是較為樂見的，而針對不同類型的消費者，以不同的方式進行銷售，就不需要將每筆行銷廣告發送給每位客戶，進而替業者省下一筆可觀的預算。因此，如何找出不同類型的電信用戶則成為本研究的目的，以現有的通信資料庫進行資料探勘，區別出各類型的用戶提供業者參考。 區別客戶類型的研究大多以個人資料及歷史消費紀錄作為研究資料來源。然而，國內電信業者對於申請用戶的資料記錄並不詳細，亦無法確定登記申請者與電話使用者為同一人。因此，本研究採用大量的歷史通訊紀錄為資料來源，應用自組織映射圖網路技術分析用戶行為特性，從中區別各類型的用戶並指出各類別特性，並整合倒傳遞網路技術進行模型建構，以期發展可靠的預測模型。With the growth of telecommunication market, every Internet Service Provider (ISP) proposes many kinds of marketing strategies to expand their market and gain customers. Although lots of these marketing plans can attract users and increase volume of sales of products, most of these strategies cost much and resource wasted for ISPs. For customized marketing, nowadays, discovering different customer communities through customer profiles and historical consumer behaviors is important and urgent, and it motivates our research. For promoting specific products to different groups, we collect a large amount of historical data of telecommunication users, and apply state-of-the-art classification methods to categorize different communities of consumer behaviors. Due to the stabilization and efficiency of neural network, we employ self-organizing map and back-propagation network algorithms to generate a prediction model, and utilize it to group different consumer communities, then analyze them from different phases for promotion. From the experimental results, we believe this approach can classify communities efficiently and accurately, and it is workable to support marketing in telecommunication market.
