隨著線上拍賣交易量的快速成長，陸續發生許多交易糾紛，其中最嚴重的莫過於詐騙。拍賣平台提供的二元名聲系統不足以保護消費者避開陷阱，有時更淪為詐騙者的行騙工具。有鑑於此，學者們紛紛提出各種詐騙偵測方法，期能協助使用者避開詐騙、安心交易。典型的做法為設計一套詐騙偵測屬性集，並使用不同的學習演算法來塑模。這些方法雖然各有特色，但很少能兼顧偵測的成本與效益，此外，相關研究多使用單一偵測模型，使其效能受到限制。為了節省偵測成本，並提升偵測準確性，本研究首先發展一套詐騙偵測屬性篩選演算法-EFCBF，期能以較少的屬性，獲得較佳的偵測結果，以降低偵測成本。根據挑選的屬性集，本研究進一步提出一套平衡式詐騙偵測流程，以互補方式結合多個偵測模型，提升總體的準確性。為了驗證提出方法的有效性，我們蒐集了Yahoo!Taiwan的交易資料進行實驗，並與前人研究比較。結果顯示EFCBF屬性挑選方法與平衡式詐騙偵測流程均能以較少的成本，提供較佳的詐騙偵測結果。最後，為了增進實用性，本研究根據上述方法，開發了一套線上拍賣決策支援工具-AuctionGuard，協助使用者挑選交易對象，及時避開詐騙與交易糾紛，提升交易的滿意度。

With the rapid growth of online auction transaction, there have been incidents of many trade disputes, the most serious of which is fraud. The binary reputation system in online auction platform is insufficient to protect consumers avoid the trap, sometimes even become a defraud tool used by fraudsters. For this reason, scholars have proposed a variety of fraud detection methods to help users to avoid fraud to help ease transactions. The typical approach for detect fraud is the design of a fraud detection attribute set, and use different learning algorithms to build fraud detection models. Although these methods have their own characteristics, but few take into account the costs and benefits of detection, in addition, related studies using a single detection model and so that performance is limited. In order to save the detection cost and improve the detection accuracy, this research have developed a fraud detection attribute filtering algorithm-EFCBF, using fewer attributes but gain better detection results to reduce the detection cost. According to the set of selected attributes, we further proposed a balanced fraud detection process, the combination of multiple detection models in a complementary manner to enhance the overall accuracy. In order to verify the proposed method, we experiment by using the collected transaction information in Yahoo! Taiwan, and compared with the previous studies. The results show that EFCBF and balanced fraud detection process both can provide better fraud detection results at less cost. Finally, in order to enhance the practicality of this research under this approach, we developed a set of online auction decision support tools-AuctionGuard, to assist the user in the selection of trading partners in a timely manner to avoid fraud and trading disputes for improve satisfaction with the transaction.

目錄
1.	緒論	1
2.	相關技術與概念介紹	5
2.1	詐騙偵測屬性	5
2.2	屬性篩選方法	10
2.3	詐騙偵測模型	13
2.3.1	偵測模型的建立	13
2.3.2	多階段詐騙偵測流程	16
3.	詐騙屬性篩選及模型建構	18
3.1	EFCBF(Enhanced FCBF)篩選方法	18
3.2	平衡式詐騙偵測流程	22
4.	系統實作	26
4.1	詐騙偵測	27
4.2	名聲評估模組	29
4.2.1	趨勢分析	30
4.2.2	綜合分析	37
5.	實驗結果	40
5.1	實驗設定	40
5.2	屬性選取演算法之效能驗證	42
5.2.1	EFCBF之效能驗證	42
5.2.2	少量資料狀況下之EFCBF效能驗證	46
5.3	平衡式詐騙偵測流程效能驗證	50
5.4	新資料詐騙偵測效能驗證	52
6.	結論及未來展望	54
參考文獻	56
附錄一、FCBF 10次實驗所挑選之指標集	61
附錄二、EFCBF 10次實驗所挑選之指標集	63
附錄三、實驗結果匯總比較	69

 
表目錄
表 2 1：Chau等人所提出之屬性集(Chau et al., 2006)	7
表 2 2：Chang & Chang提出的屬性集(Chang & Chang, 2009)	8
表 2 3：鄭孝儒所提出的詐騙偵測屬性集(鄭孝儒, 2011)	9
表 2 4：電腦公司顧客資料數據表	14
表 5 1：交易資料年份分布	41
表 5 2：Confusion Matrix	41
表 5 3：10次實驗中FCBF曾經挑出過的屬性	43
表 5 4：10次實驗中EFCBF曾經挑出過的屬性	43
表 5 5：不同屬性集詐騙偵測比較	45
表 5 6：爆發前90天詐騙偵測實驗結果	47
表 5 7：爆發前60天詐騙偵測實驗結果	48
表 5 8：爆發前30天詐騙偵測實驗結果	49
表 5 9：不同資料切割方式下之偵測成功率比較	50
表 5 10：平衡式詐騙偵測模型實驗與各項實驗比較	51
表 5 11：新加入詐騙者發生年份分布	53
表 5 12：新加入資料後詐騙偵測實驗結果	53

 
圖目錄
圖 2 1：FCBF演算法之虛擬碼(Yu et al., 2004)	13
圖 2 2：鄭孝儒(2011)所提出的二階段詐騙偵測模型	17
圖 3 1：EFCBF之虛擬碼	21
圖 3 2：平衡式詐騙偵測模型架構圖	25
圖 4 1：系統架構圖	27
圖 4 2：詐騙偵測工具，偵測為正常賣家	28
圖 4 3：詐騙偵測工具，偵測為詐騙	29
圖 4 4：賣家之交易量、交易金額與平均金額走勢圖	32
圖 4 5：拍賣趨勢分析─交易量MA7	35
圖 4 6：拍賣趨勢分析─交易量MA30	35
圖 4 7：拍賣趨勢分析─平均金額MA7	36
圖 4 8：拍賣趨勢分析─年度交易分布圓餅圖	36
圖 4 9：賣家基本資料介面	39

參考文獻
1.	王俊程, 邱垂鎮, & 葛煥元. (2005). 以交易記錄的社會網絡結構建立線上拍賣哄抬評價的偵測指標. 資訊管理學報,12(4), 143-184.
2.	洪儀玶. (民國96). 具早期預警能力之線上拍賣詐騙偵測 = An early warning system for fraud detection on electronic auction / 洪儀玶撰. 臺北縣淡水鎮 : 淡江大學資訊管理學系碩士班,.
3.	鄭孝儒. (民國100). 線上拍賣潛伏期詐騙者之有效偵測 = Effective detection for latent fraudsters in online auctions / 鄭孝儒撰. 臺北縣淡水鎮 : 淡江大學資訊管理學系碩士班,.
4.	台灣資策會，推動電子商務產業，http://www.iii.org.tw/business/2_5_7.asp，2012
5.	刑事警察局，最新犯罪手法宣導，http://www.cib.gov.tw/crime/crime02_2.aspx?no=581，2012
6.	Chau, D.H., and Faloutsos, C. (2005). Fraud detection in electronic auction. European Web Mining Forum at ECML/PKDD,
7.	Chau,D.H., Pandit,S., and Faloutsos,C.:Detecting fraudulent personalities in networks of online auctioneers. Proceedings of PKDD 2006, pp.103-114. (2006)
8.	Chau,D.H., Pandit,S., Faloutsos,C., and Wang,S..: NetProbe: A fast and scalable system for fraud detection in online auction networks. Proceedings of the 16th International Conference on World Wide Web, pp. 201-210. (2007).
9.	Chang, W.H. and Chang, J.S., An Early Fraud Detection Mechanism for Online Auctions Based on Phased Modeling. The IEEE International Workshop on Mobile Systems E-commerce and Agent Technology (MSEAT 2009), Dec 3-5, 2009, Tamkang University, Taipei, Taiwan.
10.	Chang, W.H. and Chang, J.S. "An Online Auction Fraud Screening Mechanism for Choosing Trading Partners"　2010 The 2nd International Conference on Education Technology and Computer (ICIEE 2010), June 22-24, 2010, Shanghai, China
11.	Chang, W.H. and Chang, J.S., Using Clustering Techniques to Analyze Fraudulent Behavior Changes in Online Auctions. 2010 International Conference On Networking and Information Technology (ICNIT 2010), June 11-13, 2010, Manila, Philippine, http://www.icnit.org/ (2010)
12.	Chang, W.-H. and Chang, J.-S. , “Early Fraud Detection Procedures for Online Auctions,” summated to Electronic Commerce Research and Applications, 2011. (accepted for publication subject to minor revision)
13.	Chang, W.-H. and Chang, J.-S. , “Analysis on the Strategies of Fraudulent Behavior Flipping in Online Auctions,” 2011 (Working Paper)
14.	Dellarocas, C. Immunizing online reputation reporting systems against unfair ratings and discrimatory behavior. Proceedings of the second ACM Conference on Electronic Commerce, October 2000.
15.	Dellarocas, C. Mechanisms for coping with unfair ratings and discriminatory behaviour in online reputation reporting systems. International conference on Information Systems, December 2000.
16.	Dellarocas, C., "Analyzing the Economic Efficiency of eBay-like Online Reputation Reporting Mechanisms," ACM EC'01, Oct. 14-17, 2001, pp. 171-179.
17.	Dellarocas, C. and Wood, C. A. The Sound of Silence in Online Feedback: Estimating Trading Risks in the Presence of Reporting Bias, Management Science 54(3), March 2008, 460-476.
18.	Flannery, B. P., Press, W. H., Teukolsky, S. A., & Vetterling, W. (1992). Numerical recipes in C. Press Syndicate of the University of Cambridge, New York,
19.	Goes, P., Tu, Y. and Tung, A.,:Onine Auctions Hidden Metrics", Communications of the ACM, vol. 52, No.4, pp. 147-149. (2009).
20.	Guyon, I. & Elisseeff, A., An Introduction to Variable and Feature Selection. Journal of Machine Learning Research, 3, pp.1157-82, 2003.
21.	Goes, P., Tu, Y. and Tung, A.,:Onine Auctions Hidden Metrics", Communications of the ACM, vol. 52, No.4, pp. 147-149. (2009).
22.	Quinlan, J. R..:C4.5:Programs for machine learning. San Mateo CA: Morgan Kaufmann, (1993)
23.	Witten, I. H., Frank, E., & Hall, M. A. (2011). Data mining: Practical machine learning tools and techniques Morgan Kaufmann.
24.	Yu, L., & Liu, H., “Efficient feature selection via analysis of relevance and redundancy”. The Journal of Machine Learning Research, 5, 1205-1224,2004.
25.	Yu, B., and M. P. Singh, "Detecting Deception in Reputation Management," AAMAS'03, July 14-18, 2003, pp.73-80.
26.	Yu, B., and M. P. Singh, "Searching Social Networks," AAMAS'03, July 14-18, 2003, pp. 65-72.