探討在RFM分析與協作性過濾架構下之顧客關係管理策略-以電子商務為例

Abstract

客戶關係管理(CRM)作為一種行銷理念，已提供了瞭解客戶需求和推論客戶傾向趨勢的基礎。但是，大多數企業在制定長期正向的客戶關係管理策略時，仍存有困難。 為了預測客戶行為的變化，並提供顧客導向的服務，本研究提出了基於RFM分析和協作性過濾的CRM策略。 首先，以RFM 方法和K-means方法來衡量客戶的忠誠度，收集有類似RFM加值的顧客成為群組。本研究定義了8種群組的顧客，並將這些群組 與實際資料比較，進而評估RFM值是否有助於預測客戶的購買行為。 結果顯示，RFM策略，有助於預測下次採購並且知道誰是忠誠度高的顧客。使用RFM分析，市場行銷人員可以按照RFM群組針對每個族群的進行忠誠度行銷活動。 其次，要提高顧客的滿意度，本研究提出了基於協同過濾的推薦方法。 這個推薦系統是有效的技術，它藉著給客戶推薦正確的產品，主要可以增加利潤。 本研究以兩段推薦程序方式來進行，第一頁推薦和第二推薦。 而且為了比較第一段的推薦方法，本研究提出的客戶可能採購項目類別推薦和A-線上電子商業採用的最佳暢銷產品推薦方法兩者都納入評估。 實驗結果顯示，研究提出的推薦和暢銷推薦結合的方法效果最好。因為產品的Level會影響推薦系統的效率。 從第二段推薦評估的結果，類別推薦時，TOP 6方法獲得最好的結果；產品推薦時，TOP3方法獲得最好的結果。

Customer relationship management as a marketing idea has provided understanding of the customer’s needs and delivery the philosophy of customer orientation. However, most companies have difficulty in customer relationship management strategies to develop long-term and positive relationship with customers. To predict change in customer behavior and provide customer-driven service, this study has proposed CRM strategies based on RFM analysis and collaborative filtering. First, the RFM method, Recency(R), Frequency(F) and Monetary(M), and K-means method were used to measure customers Loyalty and cluster customers into groups with similar RFM values. The 8 segments of customer were defined and were compared with real word data for evaluating whether RFM value helps to predict customer’s purchasing behavior. The result shows that RFM strategy helps marketers to predict next purchase and know who has strong loyalty. Using this analysis, marketer can plan each of loyalty programs depending on RFM groups. Second, to enhance the customer satisfaction, the recommendation method based on collaborative filtering has been proposed. The recommendation system is powerful technology mainly to promote items for increasing profit by recommending right products to customers. In this study, two-step recommendation process was conducted, first page recommendation and second recommendation. The proposed recommendation providing items and categories which customers are likely to purchase and the best selling recommendation system A-online shopping mall adopts, were evaluated in first page recommendation. The experimental results demonstrate that the combined method, proposed recommendation for item level and best selling recommendation for category level, performs better than the existing recommendation because the efficiency of recommendation is affected by the level of the taxonomy. From the result of second recommendation experiment, the recommendation of category has best performance under TOP 6 and the recommendation of item has best performance under TOP 3.