Comparing paper level classifications across different methods and systems: an investigation of Nature publications

• Published in: Scientometrics Vol. 127; no. 12; pp. 7633 - 7651
• Main Authors: Zhang, Lin, Sun, Beibei, Shu, Fei, Huang, Ying
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.12.2022; Springer Nature B.V
• Subjects: Classification; Computer Science; Information Storage and Retrieval; Library Science; Machine learning; Marking and tracking techniques; Research management; Scientific papers; Scientometrics; Paper level classifications; Fields of research (FoR); WoS subject categories; Nature subjects; InCites; Dimensions; Springer Nature
• ISSN: 0138-9130; 1588-2861
• DOI: 10.1007/s11192-022-04352-3

The classification of scientific literature into appropriate disciplines is an essential precondition of valid scientometric analysis and significant to the practice of research assessment. In this paper, we compared the classification of publications in Nature based on three different approaches across three different systems. These were: Web of Science (WoS) subject categories (SCs) provided by InCites, which are based on the disciplinary affiliation of the majority of a paper’s references; Fields of Research (FoR) classification provided by Dimensions, which are derived from machine learning techniques; and subjects classification provided by Springer Nature, which are based on author-selected subject terms in the publisher’s tagging system. The results show, first, that the single category assignment in InCites is not appropriate for a large number of papers. Second, only 27% of papers share the same fields between FoR classification in Dimensions and subjects classification in Springer Nature, revealing great inconsistencies between these machine-determined versus human-judged approaches. Being aware of the characteristics and limitations of the ways we categorize research publications is important to research management.