線上拍賣的蓬勃發展有目共睹，此種透過網路買賣物品的方式，已成為許多人生活的一部分。雖然網路有助於打破交易的時間與空間限制，但其隱匿性也讓交易者承擔更大的風險性與不確定性。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)來管理會員名聲。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名聲外，也常常受到有心人的操弄，甚至淪為詐騙者的工具。 有鑑於此，本研究以發展更合理的線上拍賣名聲模型為目標，提出了一套新的多屬性相對名聲計算方法。與前人研究不同之處，我們在計算特定會員的累計評價時，同時考量『領域相似度』、『交易金額』、『時間衰減度』與『評價信任度』等屬性，以得到更合理的名聲值。此外，本研究也提出以相對名聲而非絕對名聲來比較會員信譽，以避免高評價者一直處於絕對優勢的不合理狀況，並降低有心人操弄評價系統的動機。為驗證提出方法之有效性，我們蒐集拍賣網站的實際交易資料進行分類測試，結果顯示本研究提出的多屬性相對名聲模型能確能提供比現行二元名聲系統更精確之服務品質預測。尤其，在挑選名聲近似的交易對象時，更能獲得具參考價值的資訊。 The popularity of on-auction is obvious to all, trading goods on the auction sites has become a part of daily life of a modern citizen. Although the restriction on the trading time and location can be greatly alleviated by the network, however, a more risky and uncertain trading nature is involving in such a anonymous environment. To assist the members in choosing a proper trading partner, the auction sites provide a simple binary reputation mechanism to manage the reputation of members. However, it is proven that the binary reputation system is too simple and is hard to reflect the truly reputation. Focusing on this problem, a novel multi-attribute relative reputation calculation method is proposed in this paper to help the user selecting a proper trading partner. Different from the previous works, a relative but not a absolute reputation value is used to compare the trusted degree of two different members. Besides, to derive a more accurate trust model, the similarity and the transaction amount of commodities’ categories, along with the time-decay and opinion weight of the binary reputation value are combined to synthesize a overall reputation value. To demonstrate the effectiveness of our system (named BREP), the real transaction data in Taiwan’s ruten on-line auction site are collected to perform experiments. The results show that BREP can provide a more accurate suggestion than the binary reputation value in 10%-24% cases of tests. In particular, for two members with high binary reputation value, BREP often can differentiate them to help reputation requestor making choice.
