abstract

Since the introduction of the impact factors, scientometrics has gradually become an important topic nowadays in the evaluation of the research performances at various levels, including but not limited to, personal-level, department-level, university-level, etc. There exists hundreds of evaluation indices describing in different perspectives, but most indices are obtained via simple arithmetic, from a data subset, and without statistical verification. A standard practice of obtaining these evaluation indices is to analyze the big data from the citation database. This talk introduces new techniques from statistics and big data to analyze a complete citation database called the Web of Science. In specific, it consists of three parts. The first part introduces a new centrality-based evaluation index, namely the network inc articles among re-search communities in the citation network. The second part proposes a two-step approach via deep learning and regression to quantitatively describe the relationship among subjects found in the big database. The third part applies the network clustering technique to detect potential research communities in the coauthorship network, and the evolution of these communities over years are also observed. This is a collaborative project with the research metrics group in the Institute of Statistical Mathematics （ISM） of Japan.