摘要
在线评分系统中的恶意或随机打分为准确评价在线用户声誉带来了极大的挑战.对3种基于迭代的经典在线用户声誉评价算法的鲁棒性进行了细致研究.实验先将不同数量用户打分随机化,再以均方根误差为指标衡量其余用户声誉值受影响程度.实验共在3个数据集中进行,在MovieLens和Netflix两个经典实证数据集上的实验结果表明：系统中1%~60%的用户进行随机打分时,基于关联分析的CR算法始终保持很好的鲁棒性;基于打分迭代的IARR算法的均方根误差略有增大,最大值达到0.22,但整体波动较小;而改进的基于打分迭代的IARR2算法的均方根误差最大值达到0.695,其鲁棒性的较大波动是因算法受高声誉用户的影响较大.在Douban数据集上的结果表明：在打分数据稀疏情况下,CR算法也能保持很好的鲁棒性.
Malicious and spam actions in online rating systems affect the user reputation measurements greatly. By setting different number of spammers and evaluating its effect by the root-mean-square error （RMSE）, the robustness of three typical iterative-oriented online user reputation measurements was investigated. The results for MovieLens and Netflix data sets show that when facing with 1% -60% of spammers in the network, the CR algorithm has the best performance of robustness. The largest RMSE value of the iterative algorithm of reputation IARR reaches 0.22,with slight fluctuation of the RMSE. And the RMSE value of the improved iterative algorithm of reputation IARR2 reaches 0. 695. The result for Douban data set shows that the CR algorithm still maintains great robustness even when the users rate few common items.
出处
《上海理工大学学报》
CAS
北大核心
2016年第4期362-366,共5页
Journal of University of Shanghai For Science and Technology
基金
国家自然科学基金资助项目(71271126
71374177
61361125)
教育部博士点基金资助项目(20120078110002)
上海市东方学者特聘教授
上海市曙光学者(14SG42)
关键词
在线打分系统
用户声誉
鲁棒性
online rating system
user reputation
robustness
