科研项目布局差异对比方法研究——以NSF和EU FP项目为例  

Research on Comparison Method of Scientific Funding Layout ——Take NSF and EU FP Grants for Instance

作　　者：陈挺[1] 韩涛[1] 李泽霞[1] 李国鹏[1] 王小梅[1] 

机构地区：[1]中国科学院文献情报中心,北京100190

出　　处：《现代图书情报技术》2015年第7期89-96,共8页New Technology of Library and Information Service

基　　金：国家自然科学基金项目"科学结构特征及其演化动力学分析方法与应用研究"(项目编号:71173211)的研究成果之一

摘　　要：【目的】弥补目前科研项目布局分析往往局限在单一资助机构简单数量统计上的不足,从内容上提高资助机构布局差异的揭示能力。【方法】针对多源项目数据,提出一种基于文本K-means++聚类算法的分析方法,尝试从科研项目内容揭示不同资助机构的资助方向和布局重点,分析比较各资助机构在研究方向上的资助差异。【结果】使用美国NSF与欧盟FP资助的项目信息进行方法验证与案例分析,发现相对于多个关键词,基于单个关键词构建的文本特征空间有更好的聚类效果。进一步去除项目申请书摘要中项目背景、未来影响等干扰信息,只保留研究内容、研究方法等实质性研究描述文本,K-means++算法的聚类效果有进一步提升。【局限】数据清洗尚不能完全自动实现,聚类参数的预设与调整也需人工参与。【结论】实验与案例证明该方法是可行的,分析结果能够比较直观地反映资助机构的布局差异,对科研管理与决策者审视宏观科研布局、前瞻科技发展方向起到一定的辅助作用。[Objective] This study targets to improve the ability to discover research layout of funding agencies based on their funding application documents. [Methods] The K-means＋＋ clustering method is proposed to analyse the funding direction and main focus based on multiple sources of funding application documents. [Results] After validation and a case study based on the funding application abstracts from NSF and FP, it is discovered that single-word feature is more accurate than multi-word feature in the K-means＋＋ clustering. If only keep the essential contents of application abstrcts as analysis documents, the accuracy of the K-means＋＋ clustering is significantly improved. [Limitations] Data cleaning of the funding application documents is not fully automated. Adjustment of clustering parameters need to be manually controlled. [Conclusions] The K-means＋＋ clustering of fimding application documents is a practicable method by validation and case study. Research layout differences in agencies＇ funding trends are discovered and could be helpful for scientific management and policy decision.

关 键 词：科研项目 科研布局 K-means++聚类 轮廓系数 

分 类 号：G311[文化科学]