兼容异构数据的稳定评估模型

Stable Evaluation Model Compatible with Heterogeneous Data

随着我国科技实力的不断提升,各类评价活动以及参与申报奖励的项目数量和质量都在稳步提升.为了缓解专家评委的打分压力,现有项目评价机制往往采用先分组再综合的评价方式.然而,在将不同分组中的项目进行统一排序时,不同分组的组间评价差异为统一排序带来了新的挑战.基于分组评价与统一排序的矛盾,本文设计了一种提高平行分组评价公平性的稳定评估模型,用以帮助不同分组专家打分产生的异构数据可以统一排名.该算法使用归一化方法消除不同小组专家在彼此独立场景下进行打分产生的组间差异.基于反复改进原理,实现项目分数和专家权重的互评,进而求得专家的稳定权重值来消除组内专家之间因评价标准不同而导致的差异.最终使处理后的数据可以用于整体排序.

With the continuous improvement of scientific and technological strength,various evaluation activities,as well as the number and quality of projects participating in the application of awards,are steadily increasing.To relieve the scoring pressure of the expert judges Jthe existing project evaluation mechanism often adopts the evaluation method of grouping and then comprehensively sorting.However,when these projects in different groups are sorted in a unified manner,the evaluation differences between different groups bring new challenges.Based on the contradiction between group-based evaluation and unified ranking,in this paper,we design a stable evaluation model to improve the fairness of parallel group-based evaluation,so that the heterogeneous data generated by experts in different groups can be ranked uniformly.The algorithm utilizes a normalization method to eliminate the differences between groups caused by different experts scoring in independent scenarios.Based on the principle of repeated improvement,the project scores and the expert weights are evaluated mutually to obtain a stable weight value of the experts,eliminating the differences caused by the different evaluation standards among the experts.Finally,the processed data can be used for overall ranking between different groups.