由於網路拍賣的龐大使用者與低進入成本，引來了許多拍賣詐騙的情況，根據美國IC3從2003年到2011年的報告，我們可以發現網路犯罪的申訴案件數與受害人數呈現年年俱增的現象，而網路拍賣詐騙則是其中一項日益嚴重的網路犯罪項目。針對網路拍賣詐騙問題，先前研究曾提出以馬可夫鏈與類神經網路等方法之詐騙偵測方案，其偵測的效果雖然較傳統的評價系統好，但仍有改善的空間，而其產出之結果並不具可讀性，無法讓人輕易了解與使用，再者，先前研究皆未清楚說明網路拍賣資料應如何有效率的擷取，為了改善上述網路拍賣詐騙偵測的問題，本研究提出一套具有效率的網路拍賣資料擷取方法並結合多種分析工具來偵測拍賣詐騙者，最後產出可讀性高的辨識規則，以幫助拍賣使用者可以更輕易且有效的辨識較危險的賣家。在本研究中，我們將以社會網路分析、犯罪經濟學理論、賣家密度觀點來產生賣家的特徵，然後再將這些特徵導入模糊控制系統中形成模糊控制規則，再以遺傳演算法來進行模糊規則庫的最佳化學習，以建立最佳的網路拍賣詐騙者偵測規則。在實做的實驗中，我們以實際的拍賣網站"露天拍賣"作為資料蒐集的對象，並蒐集真實的拍賣交易資料以進行詐騙者監測的實驗。實驗結果發現，本研究所提出的網路拍賣資料收集機制可穩定且有效率的擷取拍賣交易資訊與使用者帳號資料，詐騙者偵測系統架構可將偵測指標降至四個，並且達到比先前研究更好的偵測效率，產出具可讀性之偵測規則以利使用，期望這些結果可以有助於使用者在辨識拍賣詐騙者上得到有效的協助。Due to the huge amount of users and low entrance cost of online auction, there are a lot of online fraud cases in online auction sites. According to the IC3 reports from 2003 to 2011, we can understand the fraud cases and victims are increasing rapidly year by year. To solve the problems of online auction fraudster detection, the previous researches develop several solutions which use Markov Chain and Artificial Neural Network techniques can increase the performance of fraudster detection. Although these solutions have better performance of detection, the output models are not readable and not easy to use for every online auction users. In addition, the previous researches didn't discuss how to collect online auction data efficiently. To improve the defects of previous solutions, this research will propose a parallel data collection mechanism and a hybrid approach which can produce the readable detection rules to detect the fraudster accounts to help the users to identify dangerous sellers. In this research, we use social network analysis to produce the behavior features and transform these features into fuzzy rules which can represent the detection rules. Then optimize the fuzzy rules by genetic algorithms to build the auction fraud detection model. For implementation, we collect the real auction data from the online auction site "http://www.ruten.com.tw" which is the most popular auction site in Taiwan. Finally, we found the proposed parallel data collection mechanism can collect and capture online auction data with account information efficiently. The fraudster detection mechanism use only four features to provide better detection performances than previous researches. The readable detection rules produced and provide more usability for online auction users. We hope the results of this research can help the online auction users to detect the possible fraudsters easier in online auction.
