基於RFM分析法之顧客適性化產品推薦機制

An Adaptive Product Recommendation System Based on RFM Method

鄭婕妤(Chieh-Yu Cheng)

指導教授 ： 李麗華

朝陽科技大學/資訊學院/資訊管理系/碩士(2010年)

摘要

企業為求能在競爭激烈的市場上取得生存之道並保有競爭優勢，因此過去以企業利益為主的經營觀念漸漸轉變為以顧客為導向的經營理念。為了幫助顧客能快速又有效地搜尋出符合需求的資訊或服務，並協助企業了解顧客喜好以滿足其需求，故必須仰賴推薦系統來分析顧客的消費行為、需求與偏好，藉此提供符合顧客個人化之產品或服務。過去有研究[4]將RFM分析法結合推薦系統提出適時性(Timely)推薦機制，此方法是分析顧客進入賣場或網站時，對過去已購買過之產品是否已達到被需求之天數，由於這個方法並非以產品被需求的區間做考量，當產品遇到重覆週期的天數時，系統判別大於週期天數仍會予以推薦，因此容易造成重覆推薦的結果因而降低推薦效率。爾後，更有Li等學者[23]欲改善顧客之個人化消費習性與其對產品週期性的需求方式，提出一套具有消費性產品被購買週期的推薦系統，以計算產品被購買週期的最小與最大天數作為推薦區間，在顧客對產品產生需求時間之前，主動推薦顧客該類產品，不過此研究主要是針對經常性消費的高忠誠度顧客予以分析並推薦，忽略了低忠誠度顧客對產品可能有興趣但卻未購買的潛在偏好之重要性。因此若能準確掌握高忠誠度顧客之產品需求，同時也發掘低忠誠度顧客之潛在產品偏好，如此便能有效提升推薦效能。
有鑑於此，本研究考慮產品購買週期與顧客消費特性，提出一套植基於RFM分析法之顧客適性化產品推薦機制，針對不同消費特性的高、低忠誠度顧客，運用RFM分析法、自組織映射圖網路、推薦系統等方法，並依據顧客的歷史交易紀錄加以分析後，於適切的時間內推薦顧客符合產品購買週期與偏好之適性化推薦列表。由實驗結果顯示，本研究所提之推薦機制，其低忠誠度顧客之推薦效能於準確率及F1指標皆優於傳統的內容導向式與協同過濾式推薦，且有超過一半以上的顧客其推薦成功率達50%以上，顯示本研究之推薦機制能提供更有效益的產品推薦。

關鍵字

推薦系統 ； 自組織映射圖網路 ； 適性化 ； RFM分析法

並列摘要

In order to maintain the competitiveness and retain the promising business development in the competitive market, the focus of marketing gradually converts from revenue-based into customer-based. To service customers, recommendation systems (RS) are commonly used to analyze the consumption behavior, customer requirements, and preference, so that the suitable personalized products or services are provided. It is known that recommendation systems can effectively provide the information or service that fits customer’s need.
In the past, a type of recommendation, called timely recommendation, was proposed which combined the RFM analysis into RS. The timely recommendation is to analyze when the purchased products will be required again, so that the product can be recommended in time. However, this approach did not take the purchase periodicity into account. In this case, it will produce redundant recommendation and, hence, could decrease the recommendation performance.
To remedy this problem, Li et al. [23] had proposed an enhanced recommendation system by analyzing product consumption period. This approach uses the product purchased period by finding the minimum and maximum range of product periodicity. The RS will actively recommend products before the customer querying this product. But, this research only focused on the customer with high-loyalty purchasing history, it ignores the potential importance of low-loyalty customer. This research suggests that the recommendation performance can be improved if the high-loyalty customers are analyzed and the potential product preference of low-loyalty customers is also studied. In view of this, this study considers the product purchasing period and the consumption characteristics of customer. As a result, an adaptive product recommendation system is proposed, which utilize the RFM method to handle both high-loyalty customer and low-loyalty customer.
According to the various consumption characteristics for both high-loyalty and low-loyalty customer, the RFM analysis and the Self-organization Map (SOM) network are used to analyze customer based on their historical transaction records. The recommendation is made when the period of product purchasing is matched.
To prove the adaptive product recommendation system has better outcome, the experimental results show that the recommendation performance of the proposed approach is better than the traditional content-based and collaborative filtering recommendation, in terms of precision and F1 measure. In addition, the recommendation successful rate of the proposed system is more than 50%, which indicates the proposed method can provide effective recommendation.

並列關鍵字

RFM ； Adaptive ； Self-Organizing Map (SOM) ； Recommendation Systems (RS)

參考文獻

[1] 邱宏彬、蘇建源(2004)，「一個可彈性支援顧客關係管理與資料庫行銷之模糊RFM module」，電子商務學報，第六卷，第149-174頁。

[7] M. Balabanovic and Y. Shoham(1997), “Fab: Content-based, Collaborative Recommendation,” Communication of ACM, vol. 40, no. 3, pp. 66-72.

[8] K. W. Cheung, J. T. Kwok, M. H. Law, and K. C. Tsui(2003), “Mining Customer Product Ratings for Personalized Marketing,” Decision Support Systems, vol. 35, no. 2, pp. 231-243.

[9] Y. B. Cho, Y. H. Cho, and S. H. Kim(2005), “Mining Changes in Customer Buying Behavior for Collaborative Recommendations,” Expert Systems with Applications, vol. 28, no. 2, pp. 359-369.

[10] Y. H. Cho and J. K. Kim(2004), “Application of Web Usage Mining and Product Taxonomy to Collaborative Recommendations in E-Commerce,” Expert Systems with Applications, vol. 26, no. 2, pp. 233-246.