Larimel, J.C; Lareau, F et Malaterre, C
(2023).
« La méthode de modélisation thématique CFMf basée sur le clustering neuronal avec maximisation des traits: Comparaison avec LDA sur des études scientifiques. ».
Rencontres de la Société Francophone de Classification, pp. 67-72.
Fichier(s) associé(s) à ce document :
Résumé
The improvement of topic modeling methods remains a major concern for unsupervised analysis of textual data. We propose here a topic modeling approach based on neural clustering and feature maximization. We compare its performance to that of LDA by applying both methods to a large reference corpus of full-text philosophy of science articles. The results show very significant improvements in key quantitative performance measures such as coherence, as well as qualitative results.