在電子商務中透過資訊網路整合的信任模型

Abstract

在電子商務和其他網際網路的應用中，網際網路、甚至社交網路的使用者之間的信任關係有著重要的作用。然而，建立信任精確的數學模型，亦或在電子商務系統上以信任為基礎開發出的應用一直沒有得到令人滿意結果。在本文中，我們提出了一個以機率理論為基礎的框架，定量的描述信任為一數學推理量，進而模擬在電子商務系統中，消費者和銷售商的之間的信任行為模式。我們首先討論網際網路和社交網路上信任的性質。然後，我們構造電子商務系統的拓撲並應用統計推論得出更可靠的信任值。進而開發出一個可以有效抵抗惡意行為的賣家的可靠演算法。通過社交網絡上的學習及分散決策的概念，我們提出了可以保持信任估計的準確性且同時更好地防止潛在的惡意行為的演算法。模擬的結果顯現我們所提出的方案在電子商務中具有良好的估計準確度，在面臨的一個惡意用戶時能夠並保持強勁的性能水平。此外，以一個相當有系統的方式來討論從不同的網站收集來帶有不確定性的資訊的信任衡量。透過收集更多的資料，也提出一個連續檢測方法以提供更好的性能。模擬的結果也顯示在實際上運用研究結果與設計現實世界中的應用程式的可能性。

Trust among Internet users and thus social networks plays an important role in e-commerce and other Internet applications. However, the precise mathematical model of trust and thus applications based on trust in e-commerce system has not been satisfactorily established yet. In this paper, we present a probability theoretic framework to quantitatively measure trust as mathematical reasoning and to model the behaviors of consumers and sellers in the e-commerce system based on trust measure. We rst summarize properties of trust in Internet users and their social networking. Then we construct the topology of e-commerce system and apply the statistical inference to derive more reliable trust measure. A reliable algorithm, which is robust to malicious behaviors of the sellers, is therefore developed. Via social network learning, distributed decision is proposed to maintain the accuracy of trust estimation and to better against potential malicious behaviors. Simulations demonstrate that our proposed scheme shows good accuracy in estimation of confi dence level and retains robust performance facing a number of malicious users in the e-commerce system. Besides, trust measure with uncertain information, usually collecting from di erent websites, is develop in a systematic way. A sequential detection approach is also proposed to better o the decision performance by collecting more data. Simulations also show a possibility for practical application designing with real world data.