RulerRep:一种基于偏离度的过滤不实评价新方法

RulerRep:Filtering out Inaccurate Ratings in Reputation Systems Based on Departure Degree

信誉系统在解决开放网络环境中的信任问题时,较传统技术具有明显优势.然而不实评价的存在严重降低了信誉系统的可用性.文中提出一种基于偏离度的不实评价过滤方法:RulerRep.该方法以用户自身与服务提供者的直接经验为标尺,度量评价者的评价准确性,并以该评价准确性定义评价者的平均偏离度.在融合多个评价以计算服务提供者信誉的过程中,用该平均偏离度导出其评价的权重,使得平均偏离度大的节点的意见权重较小,从而达到过滤不实评价的效果.最后,以平均均方误差为指标,在实验仿真中与TRAVOS算法和没有使用过滤技术的Beta Reputation系统进行了性能对比.实验结果表明,在50%与100%的评价者均为恶意节点的情况下,RulerRep仍显示出接近理论最优过滤的性能,并大幅优于同类技术.

Compared with conventional techniques, reputation systems are more capable of dealing with trust in an open environment. But unfair ratings in reputation systems would slow down the system＇s availability. To filter out unfair ratings, this paper proposes a new model, Ruler- Rep. In this model, to measure the accuracy of a rator＇s history ratings, the authors use the service requester＇s interaction experience with other service providers as a rule, which is defined as the rator＇s departure degree. Then, when ratings from all rators are merged to calculate an unfamiliar service provider＇s reputation, each rating＇s weight is derived from the source rator＇s departure degree, namely the weight would be small if corresponding rator has a big departure degree. By using this method, the impact of the unfair ratings can be minimized. In simulations, RulerRep is compared with Beta Reputation system（with no unfair-rating-filtering techniques） and TRAVOS. The results show that, in environments where 50% and 100% of the rators are malicious rators, RulerRep has a good performance which is very close to that of the theoretical best filtering and also much better than that of using similar unfair-rating-filtering techniques.