Knowledge discovery for dynamic textual data: temporal patterns of topics and word clusters in corpora of scientific literature

The study aims at comparing two methods for tracing the temporal evolution of topics and keywords in corpora of scientific literature: the well-known Latent Dirichelet Allocation and a new knowledge-based system that has been developed in a functional data analysis unsupervised perspective. Object of the study is a corpus of abstracts of articles published by the American Journal of Sociology over a century (1921-2018). Our study advocates that the two methods might not be seen as alternative but rather as integrable means to improve the interpretation of findings.