摘要： 科技评价活动中往往存在不可靠的评审数据，直接用这些数据反评估专家的水平可能会导致误差甚至得出错误的结果。为解决该问题，根据不可靠数据只会分布于评审数据两端的特点，使用格鲁布斯测试法检测评审数据中的异常点，采用欧氏距离作为评估指标，再结合累计数、命中率、成功率等指标，确定专家的评审水平。实验结果证明，该方法得到的评估值更可靠。

关键词: 科技评价, 反评估, 数据预处理, 异常点检测, 格鲁布斯测试法, 同行评议

Abstract: Unreliable data may appear in the peer review process of science and technology evaluation. It may draw the wrong results if using these data directly when evaluating experts’ capability. In order to solve this problem, according to the phenomenon that the unreliable data can only be distributed in both ends of the assessment data, this paper applies Grubbs’ test method to detect the outliers, takes Euclidean distance as index, and considers other indexes to realize anti-assessment. Experimental results demonstrate the effectiveness of the method.

Key words: science and technology evaluation, anti-assessment, data preprocessing, outlier detection, Grubbs’ test method, peer review

中图分类号: 

• TP391