基于评价者过滤的个性化信任模型

信任问题是困扰C2C电子商务发展的重要瓶颈.在传统模型基础上,本文进一步引入5梯度的连续性评分标度、交易次数、多维评分、惩罚因子四项因素,研究两种商品交易历史对卖家信任的影响.为满足个性化选择,创建了评价者过滤机制,在此基础上构建可满足用户需要、防范信用欺诈的个性化信任模型,通过仿真实验验证模型的有效性.

A Personalized Trust Model Based on Filtering Evaluators

E-commerce in the Consumer-to-Consumer(C2C) area has received much attention from the academics and industry because of the acceleration of cyberization and the fast growth of Internet users in China. In 2009 C2C market generated MYM225 billion dollars in sales revenues,with a growth rate of more than 100%.However,the information asymmetry between buyers and sellers increases transactional uncertainty and risks,such as malicious attack,and Internet fraud.Meanwhile,buyers have different consumption preference and security requirements for doing transactions on the Internet.These emerging issues have made the credibility issues and personalized recommendation mechanisms new research topics in the field of e-commerce Theoretical studies have shown that online trust system can effectively solve these two primary problems,and the trust model is a key determinant for the effectiveness of an online trust system.Therefore,it is important to study the personalized trust model in the context of e-commerce because it can have great impact on e-commerce proliferation. Two assumptions are made in our proposed trust model:(1) the model consists of two types of commodities: the commodities buyers want to buy s and the other commodities ;(2) the model consists of two types of transaction histories: the direct transaction log and recommended transaction log.The transaction history is the main basis to measure trust.The direct trust and the recommended trust are calculated according to two types of transaction histories.In addition,the comprehensive trust value is calculated to measure the degree of tendency that the buyers would transact s with the sellers.While evaluating the direct trust,the authors not only distinguish the different effect of the two types of commodities on the model,but also propose a rating scale of 5 gradients.To increase the reliability of the model,we introduce 6 major factors to modify the model,including transaction time,transaction amount,credibility of evaluators,multi-dimensional feedback score,transaction frequency and the penalty factor. To assess the revised trust,This paper compares three similarities in factor score,weight of factor score and average score-amount through the selection mechanism of evaluators.The evaluators with low similarities(referred as invalid evaluators) are filtered and only the evaluators with high similarities(referred as effective evaluators) are retained.On the basis of the effective appraisal information,this paper makes the recommended information more accurate,enhances the precision of the comprehensive trust value,and meets the buyers’ requirements of finding the right sellers according to their individual needs. In the end,the paper tests the accuracy and the Anti-Attack capability of the trust model in four types of transaction acts by MATLB simulation experiments.The proposed model has a better effect in determining the penalty for the fraud services,preventing such malicious acts as the colluding evaluation and slander evaluation.