Field classification of publications in Dimensions: a first case study testing its reliability and validity

Dimensions is a research data infrastructure and tool, including grants, publications, citations, clinical trials, and patents in one place. An interesting feature of Dimensions is its field classification scheme, which is not based on journal classification systems, as in the Web of Science or Scopus, but on machine learning. Each publication is assigned to at least one field. Using the set of my own publications, I investigated whether they were reliably and validly assigned to fields. The results put in question the reliability and validity of the scheme. Large scale studies seem necessary to investigate the scheme in more detail.