線上拍賣的蓬勃發展有目共睹，由於不受時間與地點的限制，讓成交量得以大幅增加。雖然如此，但網路交易的隱匿特性，也相對帶給使用者極大的風險。因此，爲協助會員挑選、過濾交易對象，許多拍賣網站均提供簡易的二元名聲系統(Binary Reputation System)來管理成員名聲。但此種名聲系統設計過於簡單，除了無法真實反應交易者的實際信譽外，也常受到有心人士的操弄，以各種方式哄抬評價，製造假名聲。尤有甚者，詐騙者更以此為工具，引誘並詐取不知情的消費者。既然線上拍賣是電子商務重要的一環，若無法有效解決這些問題，將不利於其長遠發展。有鑑於此，本計畫將針對上述重點，發展有效且可行的解決方法，預計發展的方向包含: (1)在現有的二元名聲系統之下，發展具有預警能力的詐騙偵測方法，以協助使用者在交易前辨識詐騙者。此外，也希望能對詐騙類型進行分類，以利防範或因應。(2)改進目前的名聲計算系統，納入更多的考慮因素，以避免人為操作或誤判，並期待透過學習與最佳化演算法，縮短計算結果與實際名聲值之間的差異。 The popularity of online auctions is obvious to all. The amount of trading is extensively increased due to the releasing of trading time and physical locations of traders. However, the anonymous features of members in online auctions also incur the risk and uncertainty for trading. In view of this, auction sites provide simple mechanisms, such as binary reputation system, to help the users in selecting proper trading partner. But, such a simple mechanism is hard to represent the real reputation of a trader. Besides, it also easily manipulated by fraudsters. As online auctions are one of the most important members of e-commerce, the above problems should be resolved properly so that it can be kept in continuously progress. Therefore, the purposes of this project is to develop effective and efficient methods to solve the above issues, including (1) based on the existent (but weak) binary reputation system, developing effective early-warning fraud detection techniques, which can alter an fraud before it really happens, (2) improving the existing reputation calculation method by in including more relative features to provide the human users more accurate value for selecting proper trading partners.