0566673 - FLÚ 2023 RIV eng A - Abstract
Hladík, Radim - Renisio, Y.
Topic modeling and classification of scientific disciplines.
[International Conference on Science and Technology Indicators (STI 2022). From Global Indicators to Local Applications /26./. Granada, 07.09.2022-09.09.2022]
Method of presentation: Přednáška
Event organizer: University of Granada
URL events: https://sti2022.org/ 
R&D Projects: GA ČR(CZ) GJ20-01752Y
Institutional support: RVO:67985955
Keywords : topic modeling * classification * disciplines * theses * science
OECD category: Information science (social aspects)
Result website:
https://doi.org/10.5281/zenodo.6957149

This paper evaluates the possibility of classifying Ph.D. theses into disciplines by using a bottom-up empirical approach based on topic modeling. It examines a dataset of 334810 Ph.D. theses submitted at French universities between 2006 and 2020. In this comprehensive dataset, the variable “discipline” does not rely on any controlled vocabulary or disciplinary ontology. Consequently, there are 23057 unique labels for the variable of which 14538 appear only once. Such situation renders impossible any full-scale analysis of the data from the perspective of scientific disciplines. Our topic model is built atop of abstracts of 285311 of theses in French that include a title, keywords, and abstract. After applying the TopSBM algorithm, we obtained a topic model with 7 levels of hierarchy. The outcomes of our experiments with classification of theses into disciplines suggest that topics derived from purely textual data implicitly capture information about disciplines. This quality of topic modelling can be of great benefit when dealing with datasets where disciplinary information is unavailable or unreliable and where citation records are absent (as it remains the case especially in the Humanities).
Permanent Link: https://hdl.handle.net/11104/0339296