同行评议中的异常数据检测方法研究——以科研项目评审为例

Research on Abnormal Data Detection in Peer Review——An Example of Scientific Project Evaluation

同行评议是对科学技术活动及其产出进行科学评价的主要方式之一,然而评议过程中存在的异常评审数据可能会对科研项目评审结果的公平、公正性产生影响。为此,本文提出了一种用于识别专家评议异常数据的检测方法,该方法首先利用云模型中的逆向云算法计算专家评审数据的异常程度,并进行检测;然后,根据检测结果,采用基于协同过滤的异常评议数据修正算法,对专家异常评议数据进行修正;最后,设计了一种专家评议能力计算算法,为后续科研项目评审中的专家选择提供决策依据。基于同行评议真实数据的实验结果表明,本文提出的检测方法能够有效识别出专家评议的异常数据。

Currently peer review is one of the main effective methods to evaluate the scientific projects. However, existing abnormal data seriously affects the fairness and justice of the final evaluation results. In this paper, we propose a detection method for identifying experts＇ abnormal review data. Firstly, based on backward cloud algorithm we compute the scalar quantity representing the abnormal level of every review data; then, according to detection result, we adjust the abnormal data with a new one given by the collaborative filtering algorithm; and finally, we designed an algorithm to quantize the review ability of every expert concerned, which will serve well in the next peer review process. Experiments using real dataset show the effectiveness of the proposed methods.