以消費表現為基礎之顧客群集分析

A Consuming-Behavior based Clustering Algorithm

邱創政

指導教授 ： 蔡介元

元智大學/工程學院/工業工程與管理學系/碩士(2003年)

摘要

在以滿足顧客需求為競爭關鍵的時代中，良好的顧客關係管理是企業提昇競爭力的重要關鍵。企業為了能落實顧客關係管理之客製化精神，進行適當的顧客市場區隔便成為推行的首要工作。如此一來，企業便能根據每個顧客群的消費行為模式，以及其對於企業的利潤貢獻程度，提供客製化商品與服務。然而，目前企業大多是以顧客的人口統計之特徵屬性，作為顧客市場區隔之區隔變數，無法有效地讓每個顧客群內之顧客有較為相似的消費行為模式，導致衡量顧客群的利潤貢獻程度之結果有所偏差，也讓顧客關係管理的執行效益不彰。
有鑑於此，本論文提出一以顧客的消費表現為基礎之群集分析演算法，根據顧客的消費表現，包括曾經購買過哪些商品，以及花費在每個購買商品的消費金額，作為顧客市場區隔之區隔變數，以進行市場區隔。在此顧客分群方法中，本研究融入關聯法則支持度的概念，做為比較顧客間消費表現的基礎，並以此定義「顧客相似度」的衡量方法，以便能更客觀地衡量顧客之間的相似度，並且將之作為推導「尋找群內的新質心顧客」與「顧客分群結果之品質」等衡量方法的基礎。另外，本研究運用基因演算法選取能適當地代表初始顧客群質心之顧客，以得到較佳且穩定的分群結果，可改善原本K-means 演算法以隨機方式產生初始質心之顧客，導致其分群結果之品質時好時壞而不穩定的缺點。
本研究藉由一零售商所提供的顧客購物之交易資料，以實驗的方式證明本論文之顧客分群方法，確實能有效地讓每個顧客群內之顧客都有較為相似的消費行為及模式。此外，本論文以一發卡銀行之顧客刷卡消費的交易資料，探討如何運用本論文之顧客分群方法，順利地完成顧客市場區隔之作業，以幫助該銀行在實踐顧客關係管理之過程中，更能持續有效地掌握顧客所反應出的消費表現，以達成客製化之目標。

關鍵字

資料探勘 ； 群集分析 ； 消費表現 ； 基因演算法 ； 顧客關係管理

並列摘要

To satisfy customers’requirements and increase competition in serve market, it is critical for an enterprise to enhance Customer Relationship Management (CRM). Making appropriate marketing segment for customers is the primary task to fulfill a well CRM. Through segmentation, the enterprise can offer suitable products and services to every customer group according to their unique consuming behaviors or contributed profits. To distinguish customers for segmentation purpose, most enterprises/researches uses customers’ demographic attributes as clustering variables. However, this approach makes the customers in the same cluster tend to have different consuming behaviors, so that the segmentation result is usually not as satisfied as predicted.
To conquer the difficulties, this research proposes a consuming-behavior based clustering algorithm. The algorithm distinguishes customers into clusters using consuming-behavior variables such as the amount and monetary of products purchased. The algorithm adapts the concept of “support” from association rule algorithm to measure the similarity between customers. Based on the support concept, measurable methods of finding a new centroid customer in each customer cluster and calculating the quality of the clustering result are developed. In addition, to eliminate the drawback of unstable clustering quality due to random centroid selection procedure, this algorithm uses genetic algorithm (GA) to generate new centroid points for each cluster so that a more stable and better clustering result can be achieved..
The algorithm is applied to a retailer transaction datasets and found that the developed algorithm makes the customers in the same cluster having more similar consuming models. In addition, this algorithm also successfully applies to the customers’ transaction data of a credit card issuing bank to accomplish marketing segmentation. It is found that the proposed algorithm is beneficial for the bank to achieve a better and efficient CRM.

並列關鍵字

Data Mining ； Clustering Analysis ； Consuming Behaviors ； Genetic Algorithm ； Customer Relationship Management

參考文獻

2. 林錦誼，「探勘含重點項目集之關聯法則─以賣場資料為例」，碩士論文，元智大學資訊管理研究所，2001。

6. 闕嘉萱，「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」，碩士論文，元智大學工業工程研究所，2000。

7. Agrawal, R. and R. Srikant, “Fast Algorithms for Mining Association Rules,” Proc. 20th International Conf. on Very Large Databases, pp. 487-499, 1994.

8. Bass, F. M., D. J. Tigert and R. T. Lonsdale, “Market Segmentation: Group Versus Individual Behavior”, Journal of Marketing Research, Vol. 5, 1968.

10. Blattberg, R. C. and S. K. Sen, “Market Segmentation Using Models of Multidimensional Purchasing Behavior,” Journal of Marketing, Vol. 38, pp. 17-28, 1974.