Pruzhinin B.I.


The article considers the philosophical and methodological dissonance that arises in attempts to apply scientometric methods to assess the epistemological prospects of specific research areas. The traditional methods of such assessments are based on standards and guidelines that cumulate the historical experience of scientific knowledge. This experience can appear in a variety of forms, but in any case, scientists, one way or another, are guided by the qualitative methodological characteristics of scientific knowledge, correlating them with specific cognitive situations that arise in their research practices. Similarly, scientists try to rethink, as far as possible, external, inherently socio-cultural judgments about the prospects of their research work. They make similar attempts concerning quantitative citation indicators, but each time they encounter a methodological dissonance between purely external formal indicators and significant methodological awareness of real cognitive situations. This article proposes a philosophical and methodological hypothesis, which, in the author's opinion, allows us to outline ways to solve problems that arise in the context of this dissonance. The author believes that the correlation of such heterogeneous options for assessing the prospects of scientific areas can be at least partly effective due to the appeal to the principles of digital technologies' programming that underlie the «training of self-correcting machines». Machine learning is a set of methods whose characteristic feature is AI programming not for a direct solution of a problem (in our case, the task of massive collection and presentation of citation data) but the possibility of its differentiating correction based on solutions of many similar tasks. Such a correction, in our case, makes it possible to distinguish between scientometric data, taking into account the cognitive specifics of citation agents.


scientometrics; citation indices; epistemology; philosophy of science; methodology of science; machine learning.

DOI: 10.31249/scis/2022.01.09

Download text