【作者】 周光中；

【导师】 朱卫东；

【作者基本信息】 合肥工业大学 ， 管理科学与工程， 2009， 博士

【摘要】 作为国家支持基础研究的主渠道之一,科学基金在国家创新体系中发挥重要作用。科学基金立项评估是科学基金管理的核心环节,决定了政府科技投入能否得以有效的利用,关系到优秀的科研工作者能否通过良性竞争获得有限的科技资源。目前,关于科学基金项目立项评估的研究主要是沿用一般项目评估方法,尚缺乏针对科学基金项目立项评估的特点对评估过程中的具体问题所开展的理论研究。因此,选择科学基金立项评估问题为研究对象,具有重要的现实意义和理论价值。科学基金立项评估是以评审专家的评审意见为根本依据,因此评审专家的遴选和评审意见的处理是立项评估的核心。评审专家遴选需考察的因素复杂,且遴选过程存在诸多不确定性。另一方面,科学基金项目探索性强,即使同行专家也很可能存在一些不确切知晓的内容,评审意见本身通常以语言评价信息的形式给出。因此,科学基金立项评估属于特定环境下的复杂不确定性决策问题。研究中运用D-S证据理论,灰色理论、主观证据推理理论和语言评价方法对科学基金立项评估问题进行了理论上的探索。具体研究内容和创新性成果如下:1.通过系统地比较分析国内外关于证据合成规则的主要研究成果,提出证据冲突分析方法及利用冲突分析结论灵活选用证据合成策略的新思路。首先根据Jousselme距离公式进行冲突分析,使冲突程度较低的证据结成虚拟联盟,对联盟内和联盟间分别使用不同的证据合成规则。然后通过计算证据系统对不同证据的支持程度衡量证据的可靠性,从而实现对虚拟联盟的客观赋权。通过上述方法,可以有效的解决冲突证据合成问题,并在保证收敛速度的同时降低决策风险。最后通过与经典算例的比较,验证该方法的有效性和优越性。上述方法为构建评审专家遴选与评审意见处理方法奠定了理论基础。2.针对评审专家遴选因素的复杂性和不确定性,构建了基于D-S证据理论的多遴选指标层次结构模型。通过对国内外相关文献的分析,结合国内科学基金管理的实际情况,构建一套涵盖评审能力、科技信誉和评审绩效三个方面信息的专家遴选指标体系。论述了各遴选指标的分析与评价方法,设计了基于衰退系数的累计命中率和离散率两个定量指标,并在定量指标处理中构造了中心点三角白化权函数,结合灰色理论将定量数据转化为适于应用D-S证据理论处理的形式。3.在分析现有语言评价信息处理方法的基础上,根据立项评估工作的实际需要,提出了二维语言评价信息的概念及相关决策问题,并利用主观证据推理理论处理二维语言评价信息。针对科学基金立项评估中识别与处理非共识项目的难题,提出基于二维语言评价信息的评审意见共识性分析方法。为合理的度量评审意见的非共识程度,设计了二维语言评价信息的距离公式,证明了相关性质。以若干评审意见的相似关系描述评审意见间的层次结构关系,便于分析非共识的产生原因,有利于从非共识项目中选择具备资助价值的项目。针对评审专家权重信息完全未知、权重信息为实数及权重信息为二维语言评价信息三种不同情况,设计了评审意见集结方法和基于评审意见集结结果的排序与择优规则。4.为了系统的验证研究的实用性和有效性,以国家自然科学基金的通讯评审专家遴选和面上项目评审意见处理为背景进行实证研究。在以往立项评估过程的现实数据环境下,使用基于D-S证据理论的评审专家遴选模型和基于二维语言评价信息的评审意见处理方法能够取得预想效果。通过与现行评审意见处理方法和二元语义方法的比较研究,说明应用二维语言评价信息方法处理评审意见具有一定准确性优势,并验证了科学基金立项评审中要求评审专家反馈熟悉程度信息的实际意义。更多还原

【Abstract】 As the main support channel to basic research, scientific foundation has played an important role in the national innovation system. Scientific Foundation Project Selection (SFPS) concerns whether government’s investment in science and technology can be effectively utilized and outstanding scientists can obtain limited supports by healthy competition. At present, studies on SFPS mainly use the traditional projects evaluation methods, and there is little study based on the characteristics of SFPS or regarded the specific issues in the evaluation process. As a result, it is very important both for theory and practice that the specific issues of SFPS are chosen as the research object.SFPS is strictly pursuant to peer experts’opinions, so the selection of peer experts and disposal of comments from peer experts would be the key points. In the process of selecting peer experts, all relative factors are not only complex, but also provided with the characters of uncertainty. On the other hand, Science foundation projects are of strongly explorative, even if the peer experts are also likely to face the proposal beyond their awareness, so peer experts’reviews are usually given in the form of linguistic assessment information. Therefore, SFPS belongs to a decision-making issue with complexity and uncertainty in a particular environment. The author applys D-S Evidence Theory, Grey Theory, Subjective Evidential Reasoning and Linguistic Assessment Approach to complete a theoretical exploration on SFPS. The main content and innovative research results are as follows:1. The new ideas of conflict analysis and choosing evidence combination according to conflict analysis result are suggested based on the comparative discussion of relative studies at home and abroad. Via conflict analysis by Jousselme distance, evidences with less conflict would form virtual leagues, and then DS rule and ER rule are appled to combine the evidences in the same league and the evdences of different leagues respectively. To measure the reliability of evidence, the support degrees of each evidence by whole evidence system are calculated, and then objective weights of virtual leagues would be available. In such way, our combination rule would solve the problem of combining evidences with strong conflict, achieve convergence and reduce the risk in the decision-making. By a classic example, the effectiveness and superiority of the proposed combination rule is finally tested and verified. Forementioned approach is important theoretical basis to construct models for the selection of peer experts and disposal of comments.2. Considering the complicated and uncertain factors in peer expert selection, a multi-level model with selection criteria is constructed based on D-S Evidence Theory. By the way of analyzing domestic and overseas related literature and investigating practically, the author suggests an expert selection indicator system which contains three aspects: review ability, science credibility and review performance. The analysis and evaluation approach to each indicator is also expatiated. The author designs 2 quantitative criteria as damping constant based cumulative hit rate and cumulativedeviation rate, proposes central point triangle whiten function which makes quantitative criteria could be combined by D-S Evidence Theory directly.3. On the basis of analysing existed methods of linguistic assessment information and actual needs of SFPS, the author proposes a new concept of 2-dimension linguistic assessment information (2-DLAI) and corresponding decision-making issues, and expresses 2-DLAI by the frame of subjective evidential reasoning. Aiming at identifying and handling controversial projects in SFPS, a consensus analysis approach for experts’comments based on 2-DLAI is suggested. To measure disagreement degree reasonablly, the author designs a 2-DLAI distance formula, and prooves relevant properties. According to the similarity relations of experts’comments, the hierarchical structure relations of experts’comments are analyzed which facilitates to find out causes of disagreements and seek valuable ones from controversial projects. The experts’comments aggregation approaches are proposed under 3 different situations: non-weighted, real number weighted and 2-DLAI weighted respectively, two ordering rules based on comments aggregation results are also suggested.4. To verify the practicality and efficiency of the proposed methods, the author carries out empirical research on selecting peer experts and disposing comments from peer experts in the background of NSFC’s general projects. The expert selection model based on D-S Evidence Theory and the comments disposal method based on 2-DLAI all pass demonstration test with practical data. A comparative study shows that 2-DLAI method is more accurate than present method and 2-tuple linguistic method in the process of comments disposal, also illustrates the realistic significance of familiarity degree imformation from peer experts.更多还原