基于互逆强化模型和数理统计方法分析专家评分偏差问题与建议

"十二五"期间,高技术研究发展计划（863计划）作为引领科技发展的主要抓手之一,为提升中国整体科技实力和创新能力发挥了重要作用.验收评审专家有着评估课题完成水平、衡量科研成果产出价值的关键作用,其评分可靠性直接关系着计划实施成效评价的合理性.鉴于此,本文结合互逆强化模型和数理统计方法,以"十二五"863计划某领域课题验收为例,系统分析了专家评分的偏差问题并按表现将专家划分为8种偏差类型并分别给出应对建议.结果表明：该领域技术验收专家评分整体合理,多数专家均能给出可靠评分;宽严尺度虽然略有差异,但处于可接受范围.本文研究将为国家科技计划相关评审工作乃至其他科研管理活动中合理开展专家评价、精细化规范评审行为以及完善领域专家库、遴选专家评委人选提供参考.

Analyze the problems and suggestions of expert score bias based on inverse reinforcement model and mathematical statistics methods

During the 12th Five Year Plan period,National High Technology Research and Development Program of China (863 Program),as one of the main drivers of science and technology development,has provided important support for improving China''s scientific and technological strength and innovation ability.Acceptance experts play a key role in assessing the level of subject completion and measuring the value of scientific research achievements.The reliability of their scores is directly related to the rationality of 863 program implementation evaluation.Therefore,taking the subject acceptance in a certain field as an example,this paper combines an inverse reinforcement-based model and mathematical statistics methods to systematically analyze the rating bias of technical acceptance experts.Finally,these experts are divided into 8 categories according to their rating performances and corresponding suggestions are given respectively.The results show that the scores are reasonable as a whole,and most experts can give reliable scores;although there are some differences in the rating scales,they are basically in an acceptable range.This study will provide a reference for the review work related to national science and technology plan,and other scientific research management activities,in order to reasonably carry out expert evaluation,refine and standardize review behaviors,improve field expert databases,and select acceptance expert candidates.