GMDH在顧客關係管理之應用與分析─以零售業之便利商店為例

Implementation into GMDH for Customer Relationship Management—Case for Convenient Stores of Retail Trade

闕嘉萱(Jia-Shiuan Chueh)

指導教授 ： 陳雲岫

元智大學/工程學院/工業工程與管理學系/碩士(2001年)

摘要

本論文提出以資料群集處理技術(Group Method of Data Handling；GMDH)類神經網路法來建構顧客關係管理之預測模式，顧客購買之金額與頻率為我們探討之對象。建構完成之模式可做為成本評估之一參考指標。目前顧客關係管理部份仍然需要藉由有經驗的專家、學者來處理與解釋。但因影響之因素複雜，不易定出合理之模式。本研究中則利用GMDH演算法具有自組性，能利用多次自我學習方式，對眾多的輸入變數做最適當的篩選，而由資料本身來決定輸入變數與輸出變數之函數關係。因此不必對輸入變數有太多的事前假設，便可學習出最適當的模式。在模式驗証時，我們先針對217筆實際樣本進行GMDH之分析，接著根據217筆資料之變數分配情形，以模擬的方法分別產生六組不同樣本大小的資料進行顧客關係管理模式並比較其差異性。模式中考慮八種不同旳輸入變數，包含性別、年齡、所得、學歷、職業、商品滿意度、服務滿意度及環境滿意度。實驗結果發現，GMDH在購買頻率上之預測效果不會因訓練樣本的多寡而有顯著性之差別影響，然而在購買金額部份則會有資料大量時，預測效果能明顯增加。

關鍵字

資料群集處理技術(GMDH) ； 顧客關係管理 ； 類神經網路 ； 模擬

並列摘要

In this research, we propose a Customer Relationship Management (CRM) model constructed by Group Method of Data Handling (GMDH) neural network. Purchasing amount each time and frequency per week are our targets to considered in the CRM model, the well-established CRM model can be used as a reference index in the capital analysis. At present, it requires experts or scholars to manipulate CRM models, however, the complexity of the variables, it’s not easy to setup a satisfactory CRM model to use. In this thesis, we will use the merits of GMDH to establish a CRM model. The GMDH is a self-organizing method, it can find a best model between input variables and output variable by self-learning, and give the most significant factors at the end automatically. We use real data with size 217 to validate the constructed model, and then study the performance of the proposed model by simulation. The simulated data is generated according to the distributions of input variables in 217 real data. The input variables considered in the CRM model include sex, age, income, education, job, satisfactory of goods, satisfactory of service , satisfactory of environment. The simulation results show that the purchasing frequency of GMDH output is effected by data sample size, while the accuracy of purchasing amount is increasing as sample size is large enough.

並列關鍵字

Group Method of Data Handling (GMDH) ； Customer Relationship Management (CRM) ； Neural Network ； Simulation

參考文獻

12.梁瑞明，資料群集處理技術在半導體良率預測上之應用，元智大學工業工程研究所碩士論文，民國八十九年十二月。

3.S.Lu and K.H. Chon, “A New Algorithm for ARMA Model Parameter Estimation Using Group Method of Data Handling”, Bioengineering Conference, Proceedings of the IEEE 26th Annual Northeast, PP127-128, 2000.

4.Tadashi Kondao、Abihijit S.Pandya、Jacek M.Zrada, “GMDH-Type Neural Networks and their Application to the Medical Image Recognition of the Lungs”, Proceedings of the 38th SICE Annual Conference, PP1181-1186, 1999.

1.Farlow , Stanley J., “Self-Organizing Methods in Modeling，GMDH Type Algorithms”, Marcel Dekker, Inc.1984.

2.Frawiey. A., & Thearling, K.,” Increasing customer value by integrating data mining and campaign management software”, Database Management, PP.49-53,1999.