Investigating patterns of change, stability, and interaction among scientific disciplines using embeddings

Abstract Multi-disciplinary and inter-disciplinary collaboration can be an appropriate response to tackling the increasingly complex problems faced by today’s society. Scientific disciplines are not rigidly defined entities their profiles change over time. No previous study has investigated multiple disciplinarity (i.e. interaction between disciplines, whether of a multidisciplinary or interdisciplinary nature) at scale with quantitative methods, in profile This article explores dataset 21 million articles published 8400 academic journals 1990 2019 proposes new scalable data-driven approach disciplinarity. used relationship By creating vector representations (embeddings) measuring geometric closeness embeddings, analysis shows that similarity increased time, but overall size neighbourhood (the number neighbouring disciplines) decreased, pointing being more similar each other while same time displaying specialisation. We interpret this as pattern global convergence combined local Our is also able track development disciplines’ detecting those changed most period considered, treat compositional units, where relationships expressed analogy equations form Discipline 1 + 2 ? 3 . These findings help researchers, institutions organizations better understand react dynamics scientific research, support education sector designing curricula recruitment academics researchers.