基于证据推理规则的科学基金立项评估与决策研究

Science Fund Project Evaluation and Selection (SFPES) is of great practical significance to support the development of basic science in China. SFPES can be formulated as a multiple attribute group decision making problem. The uncertainty of evaluation information, the importance of criteria and experts, the limited knowledge of individuals and their partial understanding of the problem, all play an important role in the aggregation quality of decision information. How to dig and use the multi-source heterogeneous data in information system of National Natural Science Foundation of China (NSFC) to provide decision support has become an urgent problem to solve. This research systematically analyzes the logic framework of scientific fund review information, and constructs a comprehensive evaluation model of SFPES based on Evidential Reasoning (ER) rule. It presents a reliability measurement method to reflect the quality of evidence, revealing the hidden knowledge behind the rich data. Based on measurement of reliability, the aggregation method of evaluation information is modified, and it can effectively distinguish and reflect the quality of expert evaluation and thus influence the decision result. Based on idea of optimization learning, a nonlinear optimization model is used to obtain the core parameters of the proposed model based on ER rule. In addition, this research solves the representation of evaluation information, importance of criteria, importance and reliability of experts, and decision rule problems in the process of the multiple attribute group decision making systematically and offers a new theory and method for SFPES. The decision model based on ER rule is applied to the project evaluation and selection in NSFC, and the effectiveness and practicability of the proposed approaches are validated.

科学基金立项评估对支撑我国基础科学发展具有重要现实意义。基金立项评估是多指标群决策问题，其评估信息的不确定性、各个指标权重和专家权重的不同、以及专家个体对问题认识差异，都对评估意见合成的质量有重要影响。如何挖掘、利用科学基金信息系统中的多源异构信息提供决策支持成为亟需解决的问题。本研究系统分析科学基金评议信息逻辑框架，构建基于证据推理规则的科学基金项目评估决策模型。提出反映证据质量的可靠性度量方法，揭示丰富数据背后隐藏的知识并据此修正评估意见合成方法，能够有效区分和反映专家评议质量并影响决策结果；基于优化学习思想，利用非线性优化模型来获取基于证据推理规则合成时的核心参数；系统解决多指标群决策过程中的语言评价信息的表征，指标权重、专家重要性和可靠性、决策规则选择等问题，丰富了科学基金立项评估决策的理论与方法。以国家自然科学基金立项评估意见处理为例，验证本研究提出的理论和方法的有效性。

本项目围绕多源信息融合、证据推理、评估决策、绩效评价等方面开展一系列研究，并取得了一系列研究成果。依托该项目在国内外核心期刊上发表学术论文5篇，在经济科学出版社出版学术专著1部，5篇学术论文均发表在FMS国际B类期刊、JCR一区/二区的SSCI索引期刊上，其中，ESI高被引论文2篇。还有1篇英文论文在外审中。研究成果为更好的解决多源信息融合、综合战略绩效评价、价值评估、科学基金立项评估等问题提供了理论和方法支撑。项目组重视国内外交流合作，与国内外学者建立了广泛的学术联系。项目负责人由讲师晋升为副教授，并入选了浙江省高校领军人才计划。截止2022年毕业硕士生7人，与本项目研究内容存在直接关联性的研究生课题立项2项。

内容获取失败，请点击重试