摘要:电子商务的快速发展，给人们带来便利的同时，也逐渐体现出诸多问题，如责任意识淡薄，诚信缺乏等。传统的信息安全技术可以有效抵抗病毒，木马等网络攻击，但对于网上交易过程中出现的信任问题却无计可施。目前针对第三方交易平台如淘宝，卓越等的信誉体系的欺诈行为攻击使其不能真实反映买卖双方信任度。本文对目前的信誉信任模型进行了深入分析，并特别研究了模型的抗攻击能力和时效性，提出了一种基于社会网络的具有抗攻击能力和时效性的信誉与信任模型。主要的研究工作如下：

　　采用二维聚类过滤算法对欺诈数据进行排查；依靠交易节点交易后的评分建立信誉模型；通过信任传递关系建立信任模型；引入纠偏参数和时间权值，实现信誉值的客观化，信任值的主观化以及信誉和信任值时效化，为交易者提供决策帮助。

　　　经过多轮仿真实验表明，该模型能够正确客观地反映交易节点的交易行为以及存在交易关系节点之间的信任程度，对于错误推荐攻击，叛国者攻击和联合攻击具有较高的抵抗能力。

关键词： 社会网络，信任，信誉，抗攻击，时间权

Abstract:The rapid development of e-commerce, bringing convenience to people, gradually reflects many problems, such as the weak sense of responsibility and the lack of credit. Traditional information security technology can be effective against viruses, Trojans and other network attacks, but it can’t solve the problems of credit during the transactions. Currently the third party trading platform such as TAOBAO, Amazon and so on can’t really reflect the trust between buyers and sellers when it is attacked by fraud. In this paper, we have made a deep analysis of the current model of reputation and trust, especially the studies of the model’s anti-attack capability and effectiveness for a given period of time. We proposed a social network based, anti-attack, aging reputation and trust model. The main research work is as follows:

　　　Two-dimensional cluster filtering is used to get rid of cheating data. Reputation model is established by the rating after the transaction of transaction nodes. Trust model is established by the relationship of trust transformation. Correcting parameter is introduced to obtain objective reputation value and subjective trust value. Time weight parameter is introduced to make reputation value and trust value effective during a period of time, which would give dealers helps for decision.

　　　Several rounds of experiments show that the model can correctly and objectively reflect the trading behavior and the degree of trust between transaction nodes, and has a good resistant ability on mistaking recommendation attack, traitor attack and joint attack.

Keywords：social network , trust, reputation, attack resistant, time weight

本文主要研究了基于社会网络的信誉与信任模型，包括信用数据的采集及建模，模型的抗攻击能力。通过将推荐信任与社会网络相结合，使信任模型和实际社会网络中大众对信任的认识相一致；在模型中引入纠偏参数。纠偏参数的引入一方面使信用模型更为准确，另一方面提高了模型的抗攻击能力；在前人研究的基础上，提出一种欺诈数据二维过滤算法。该算法通过横向、纵向两个维度的数据分析，使欺诈数据的排查更加准确。