在台灣各項網路服務之廣告及電子商務，是入口網站主要的獲利方式，如何利用使用者在網路上留下的行為足跡，預測使用者對於廣告可能的回響程度來決定其因應的廣告行銷策略，然而使用者興趣會隨著時間變化而改變，為了掌握會員在不同時間點的興趣差異，本篇論文利用概念飄移(Concept Drift)的概念，依據不同類型會員來降低或提高這些會員過去歷史紀錄的影響程度，建構考慮時間因素之點擊興趣指數(Click Preference Index with Time factor, CPIT)，透過此模型有效鑑別不同行為之會員，精準找尋到高回訪潛力之會員。我們以某知名入口網站所提供的資訊作為實驗資料，由實驗證明CPIT 模型確實精準找尋到高回訪潛力之會員，以供入口網站的行銷人員有效益的行銷策略，進而增加獲利。

The profit of portal companies in Taiwan is generated by the online advertising and e-commerce. Effective advertising requires predicting how a user responds to an advertisement and then targeting (presenting the advertisements) to reflect the users’ favor. The behavioral target leverages historical users’ behaviors in order to select the ads which are most related to the users to display. Although we would not like to provide advertisements to customers, with the same concept, we would expect to predict return visit rates for the registered members in the specific category of the portal site. However, customers’ preferences change over time. In order to capture the concept drift, we propose a novel and simple time function to increase/decrease the weight of the old data to various members’ past behaviors. Then, we construct a member’s click preference index with time factor(CPIT) model, to effectively distinguish the different kinds of member behaviors and predict the  members’ return visit rates. The marketers of a portal  site can target the members with high return visit rates and design the corresponding marketing strategies. The experimental results with a real dataset have demonstrated that the CPIT model can be practically implemented and provide adequate results.

第 1 章緒論 ......................................................................................1
1.1 研究動機與目的.....................................................................1
1.2 研究架構.................................................................................6
第 2 章文獻探討 ..............................................................................7
2.1 顧客輪廓(Customer profile) .................................................8
2.2 概念飄移(Concept Drift) .....................................................11
第 3 章問題定義 ............................................................................15
3.1 訓練資料集– 一個月 .........................................................17
3.2 訓練資料集– 三個月 .........................................................19
第 4 章研究方法 ............................................................................21
4.1 考慮時間因素之興趣指數.................................................22
4.2 CPIT 與回訪率之關係探討...............................................25
4.3 CPIT 模型...........................................................................31
第 5 章實驗結果與探討 ................................................................34
5.1 CPIT 模型之準確性..............................................................35
5.2 CPIT 模型與CPI 比較.........................................................44
5.3 CPIT 模型與CPI 結合時間衰退函數之比較.....................50
第 6 章結論 ....................................................................................55
參考文獻 ..........................................................................................56
附錄—英文論文................................................................................59

圖目錄
圖 1 客戶輪廓建置流程.....................................8
圖 2 利用不同T0所呈現的時間函數曲線......................14
圖 3 CPI6各群組在七月的回訪率............................17
圖 4 CPI4-6各群組在七月的回訪率..........................19
圖 5 CPIT6與CPI6比較.....................................23
圖 6 5月份到7月份各月份回訪率對應關係....................25
圖 7 當月沒有來會員，5月份到7月份各月份回訪率對應關係....26
圖 8 當月有來會員，5月份到7月份各月份回訪率對應關係......26
圖 9 CPIT4三種方法結果比較...............................29
圖 10 CPIT5三種方法結果比較..............................29
圖 11 CPIT6三種方法結果比較..............................30
圖 12 CPIT模型之演算法...................................31
圖 13 CPIT4在5月份實際與預測回訪率.......................32
圖 14 CPIT5在6月份實際與預測回訪率.......................32
圖 15 CPIT6在7月份實際與預測回訪率.......................33
圖 16 8月份到11月份各月份回訪率對應關係..................35
圖 17 CPIT7在8月份實際與預測回訪率.......................36
圖 18 CPIT8在9月份實際與預測回訪率.......................36
圖 19 CPIT9在10月份實際與預測回訪率......................37
圖 20 CPIT10在11月份實際與預測回訪率.....................37
圖 21 CPIT7三種方法結果比較..............................38
圖 22 CPIT8三種方法結果比較..............................39
圖 23 CPIT9三種方法結果比較..............................39
圖 24 CPIT10三種方法結果比較.............................40
圖 25 比較各種方法8月回訪率之累計獲益圖..................41
圖 26 比較各種方法9月回訪率之累計獲益圖..................42
圖 27 比較各種方法10月回訪率之累計獲益圖.................42
圖 28 比較各種方法11月回訪率之累計獲益圖.................43
圖 29 針對7月份沒有來訪的會員比較CPIT7與CPI7.............45
圖 30 針對7月份有來訪的會員比較CPIT7與CPI7...............46
圖 31 CPIT7與CPI7比較....................................46
圖 32 CPIT與CPI比較8月回訪率之累計獲益圖.................48
圖 33 CPIT與CPI比較9月回訪率之累計獲益圖.................48
圖 34 CPIT與CPI比較10月回訪率之累計獲益圖................49
圖 35 CPIT與CPI比較11月回訪率之累計獲益圖................49
圖 36 預測八月回訪率CPIT與CPIT_Decay比較.................51
圖 37 CPIT與CPI_Decay比較8月回訪率之累計獲益圖...........52
圖 38 CPIT與CPI_Decay比較9月回訪率之累計獲益圖...........53
圖 39 CPIT與CPI_Decay比較10月回訪率之累計獲益圖..........53
圖 40 CPIT與CPI_Decay比較11月回訪率之累計獲益圖..........54

表目錄
表 1 五種不同行為之顧客過去三個月的來訪狀況觀察其下月份的回訪率..............................................4
表 2 CPI6第9群中，不同類型之會員的回訪率.................18
表 3 CPI6為0不同類型之會員在七月的回訪率.................18
表 4 CPI6在第7群，不同類型的會員的回訪狀況...............19
表 5 兩種不同行為之範例列表..............................44

[1] A. C. R. van Riel, V. Liljander, and P.Jurriens, "Exploring consumer evaluations of e-services:a portal site," International Journal of Service Industry Management, vol. 12, pp. 359-377, 2001.
[2] Taiwan Yahoo! Available: http://tw.yahoo.com/
[3] PCHOME. Available: http://www.pchome.com.tw/
[4] yam. Available: http://www.yam.com/
[5] Hinet. Available: http://www.hinet.net/
[6] Y. Chen, D. Pavlov, and J. F. Canny, "Behavioral Targeting: The Art of Scaling Up Simple Algorithms," ACM Transactions on Knowledge Discovery from Data(TKDD), vol. 4, p. 17, 2010.
[7] A. Tsymbal, "The Problem of Concept Drift: Definitions and Related Work," 2004.
[8] Y. H. Cho, J. K. Kim, and S. H. Kim, "A personalized recommender system based on web usage mining and decision tree induction," Expert Systems with Applications: An International Journal vol. 23, pp. 329-342, 2002.
[9] S. E. Middleton, N. R. Shadbolt, and D. C. D. Roure, "Ontological user profiling in recommender systems," ACM Transactions on Information Systems, vol. 22, p. 88, 2004.
[10] R. Bell, Y. Koren, and C. Volinsky, "The BellKor 2008 Solution to the Netflix Prize," 2008.
[11] Y. Koren, "Collaborative filtering with temporal dynamics," Communications of the ACM, vol. 53, pp. 89-97, 2010.
[12] L. Xiang and Q. Yang, "Time-Dependent Models in Collaborative Filtering Based Recommender System," presented at the Proceedings of the 2009 IEEE/WIC/ACM International Joint Conference on Web Intelligence and Intelligent Agent Technology - Volume 01, 2009.
[13] Y. Blanco-Fernà ¡ndez, J. J. Pazos-Arias, M. Là ³pez-Nores, A. Gil-Solla, M. Ramos-Cabrer, J. Garcà a-Duque, A. Fernà ¡ndez-Vilas, and R. P. Dà az-Redondo, "Incentivized provision of metadata, semantic reasoning and time-driven filtering: Making a puzzle of personalized e-commerce," Expert Systems with Applications: An International Journal vol. 37, pp. 61-69, 2010.
[14] S. Ma, X. Li, Y. Ding, and M. E. Orlowska, "A recommender system with interest-drifting," in Proceedings of the 8th international conference on Web information systems engineering, Nancy, France, 2007, pp. 633-642.
[15] Y. Ding and X. Li, "Time weight collaborative filtering," in In Proceedings of the 14th ACM international conference on Information and knowledge management, 2005, p. 485.
[16] M. Balabanovic and Y. Shoham, "Fab: content-based, collaborative recommendation," Communications of the ACM, vol. 40, p. 72, 1997.
[17] S. Gauch, M. Speretta, A. Chandramouli, and A. Micarelli, "User profiles for personalized information access," in The adaptive web, P. Brusilovsky, A. Kobsa, and W. Nejdl, Eds., ed Berlin, Heidelberg: Springer-Verlag, 2007, pp. 54-89.
[18] S. H. Min and I. Han, "Detection of the customer time-variant pattern for improving recommender systems," Expert Systems with Applications: An International Journal vol. 28, pp. 189-199, 2005.
[19] J. Lee, M. Podlaseck, E. Schonberg, and R. Hoch, "Visualization and analysis of clickstream data of online stores for understanding web merchandising," Data Mining and Knowledge Discovery, vol. 5, pp. 59-84, 2001.
[20] Y. H. Cho and J. K. Kim, "Application of Web usage mining and product taxonomy to collaborative recommendations in e-commerce," Expert Systems with Applications: An International Journal vol. 26, pp. 233-246, 2004.
[21] A. Albadvi and M. Shahbazi, "A hybrid recommendation technique based on product category attributes," Expert Systems with Applications: An International Journal vol. 36, pp. 11480-11488, 2009.
[22] Y. J. Park and K. N. Chang, "Individual and group behavior-based customer profile model for personalized product recommendation," Expert Systems with Applications: An International Journal vol. 36, pp. 1932-1939, 2009.
[23] P. a. Cunningham, N. Nowlan, S. J. Delany, and M. Haahr, "A Case-Based Approach to Spam Filtering that Can Track Concept Drift," presented at the In The ICCBR'03 Workshop on Long-Lived CBR Systems, 2003.
[24] M. Salganicoff, "Tolerating Concept and Sampling Shift in Lazy Learning UsingPrediction Error Context Switching," Artificial Intelligence Review - Special issue on lazy learning, vol. 11, pp. 133-155, 1997.
[25] M. Vuk and T. Curk, "ROC curve, lift chart and calibration plot," Metodoloski zvezki, vol. 3, pp. 89-108, 2006.