本研究針對現行的網路拍賣交易評價制度,運用社會網路分析的k-core分群演算法,以及詐欺手法特徵比對的觀念,建構出一個以網拍買賣家為導向的系統,可協助買賣家偵測「哄抬評價」及具有「賣家收款後未交付商品」詐欺特徵的潛伏詐欺者。系統的核心觀念是「以網拍買賣家為導向」,不同於為網拍平臺業者開發的偵測系統,買賣家可以主動且隨時地利用本系統查詢某個帳號的詐欺嫌疑程度,而系統會將帳號再細分成具有或不具有詐欺嫌疑,為系統使用者在拍賣網站的黑名單帳號外提供多一層防護;有經驗的買賣家更可以依據商品特性、本身經驗將系統預設參數進行個人化的調整,讓系統的詐欺偵測規則更貼近他的需求。 系統利用k-core分群演算法分析「交易網路結構」,找出隱藏其中的可疑內聚子群體(cohesive subgroup),然後分析「交易記錄」做進一步的確認,是一種運用交易網路結構與交易紀錄兩種不同資料類型的哄抬評價偵測方式;而系統對於另一個詐欺手法「賣家收款後未交付商品」的偵測方式,則是挑出六種具針對性的詐欺偵測指標來進行偵測,秉持的觀念是不同的詐欺手法其偵測指標、指標參數、及門檻值設定亦應有所不同,藉由高針對性的指標組合設定,可有效提升系統準確率。 The research focuses on existing reputation system of the internet auction and an user-oriented internet auction fraud detection system is developed by using k-core clustering algorithm and fraud rules matching. The system helps users to detect fraudsters who may inflate his reputation or receive payment without delivering goods. The core concept of the system is user-oriented, and is very different from the dealer-oriented systems. The fraud degree of some accounts can be queried actively. The accounts can be classified according fraud or not fraud and another protection besides the blacklist will be provided. The detection indices of the system can be adjusted based on the property of goods or an experienced user. Then system will be more fitted. The k-core clustering algorithm is used to analyze the structure of transaction network and the suspicious cohesive subgroup will be found out. Then transaction records will be analyzed for verification. This is an inflated reputation detection method which combines structure of transaction network and transaction records. As for the fraud of receiving payment without delivering goods, six fraud detection indices is used to detect it. The core concept is each fraud method should have corresponding detection indices and thresholds. The accurate rate can be improved by using high correlation indices.