高校杰出学者知识创新绩效的影响因素研究

    利用机器学习发现影响学者知识创新绩效的复杂非线性特征组合，能提升学者绩效和促进资源优化配置。以2016—2018年国家自然科学基金委公布的1409位国家杰出青年科学基金项目和优秀青年科学基金项目获得者为研究对象，运用学者获得项目前在web of science 刊载的14 819篇论文构建学者科研合著网络(个体网) 927个。考虑知识创新绩效的滞后性，采用学者获得项目后发表的20 824篇论文及其91 968篇被引论文，结合中国科学院文献情报中心发布的期刊分区表等多源异构数据，使用K-Means聚类将学者进行群组划分，得到特征均衡型、环境驱动型和合作创造型三类学者群组，运用决策树CART算法挖掘不同类型学者知识创新绩效的潜在决策规则。研究结果表明：(1) 知识创造具有普适性，它是促使不同类型杰出学者达成高知识创新绩效的关键因素；(2) 杰出学者应根据内外部综合条件配置科研合作关系资源，尽量避免封闭式发展路径与合著者过多导致的“规模不经济”；(3) 在不同类型学者群组中，存在影响知识创新绩效的不同特征组合，为杰出学者达成高绩效目标提供个性化发展策略。

A study of the influence factors of knowledge innovation performanceof distinguished scholars in colleges and universities

   With the development of knowledge economy becoming increasingly complicated and diversified, it is of vital significance to excavate key factors that influence scholars′ knowledge innovation performance to promote knowledge exchange, knowledge updating and innovation output among scholars. However, existing research mainly explores the complex relationship structure between the factors and innovation performance from the contingency perspective and studies the simple linear or nonlinear relationship between the variable and the innovation performance, failing to reveal the "black box" effect between variable and knowledge innovation performance and lacking systematic thought to consider the multi-level factors on the knowledge innovation performance of distinguished scholars. In order to better analyze the influencing factors of the knowledge innovation performance of distinguished scholars, this study attempts to solve the following questions: (1) how to sort out the driving factors of the knowledge innovation performance of distinguished scholars from multiple levels, and whether distinguished scholars can be divided into different groups according to the characteristic variables for focused analysis? (2) what is the complex relationship structure between internal characteristic variables of heterogeneous scholars and knowledge innovation performance, and what combination of characteristics can promote high-level knowledge innovation performance? (3) how should distinguished scholars reasonably allocate research cooperative resources and external environmental resources according to their own knowledge producing ability to improve the knowledge innovation performance? Therefore, taking data-driven knowledge innovation performance as the research objective, this paper explores characteristic factors that influence distinguished scholars to achieve high performance goals and excavates the nonlinear complex relationship structure in the rule path, laying a foundation for the implementation of well-designed performance incentive measures.Using machine learning method to study the influencing factors of knowledge innovation performance of distinguished scholars, this paper opens the "black box" of multi-level factors affecting knowledge innovation performance, and discloses the complex relationship between characteristic variables and knowledge innovation performance. With 1409 winners of the National Science Foundation for Distinguished Young Scholars and Excellent Young Scholars Published by the National Natural Science Foundation of China from 2016 to 2018 as research objects, 927 research co-author networks (individual networks) were constructed by using 14819 papers published on the Web of Science before the scholars obtained the projects. Considering the lag of knowledge innovation performance, 20824 papers published by scholars after obtaining the project and 91968 cited papers were adopted, and combined with multi-source heterogeneous data such as journal partition table published by National Science Library of Chinese Academy of Sciences, K-Means clustering was used to classify scholars into groups. Three types of heterogeneous scholar groups, namely characteristic equilibrium, environment driven and cooperative creation, are obtained, and the potential decision rules of knowledge innovation performance of different types of scholars are mined by decision tree CART algorithm. The results show that: (1) Knowledge producing is universal, and it is the key factor for different types of distinguished scholars to achieve high knowledge innovation performance; (2) Distinguished scholars should allocate research partnership resources according to internal and external conditions to avoid "diseconomies of scale" caused by closed development paths and too many coauthors. (3) In different groups of scholars, there are different combinations of characteristics that affect knowledge innovation performance, providing personalized development strategies for distinguished scholars to achieve high performance goals.Related research conclusions bring valuable management implications about how to promote high quality knowledge production: scholars with high knowledge producing ability has stronger inclusiveness on the number of partners and can also generate new ideas, new knowledge and new methods in the long-term and stable cooperation, so it is recommended that low knowledge creation ability scholars try to avoid close cooperation development path; while the administrative departments should avoid falling into the symmetric causal relationship between variables from a general perspective when promoting scholars′ scientific research cooperation, and rationally allocate the resources of scientific research cooperation according to the quality of academic environment and knowledge producing ability of scholars. In addition, some innovations have also been achieved through research. Cluster analysis and decision tree model are used to explore the influencing factors of knowledge innovation performance of distinguished scholars, and the characteristic indexes are measured based on literature to make up for the shortcomings of traditional regression analysis. Moreover, scholars with similar characteristics are divided into groups and specific categories are analyzed to make the research results more targeted.