Mapping of scientific knowledge: modeling of the graduate program in Information Science of the Federal University of Minas Gerais

Abstract

The use of computational tools has been increasingly required to organize, retrieve and understand the growing volume of data. Scientific communication has contributed both formally and informally to this phenomenon. However, managing and organizing a large collection of documents may become humanly impossible, and refutable when done manually. Topic modeling through machine learning algorithms has made it possible to organize and summarize data corpora. This study aims to identify the topics of the theses and dissertations by the graduate program in Information Science of the Federal University of Minas Gerais, southeastern Brazil (Programa de Pós-Graduação em Ciência da Informação da Universidade Federal de Minas Gerais). The main goal is to identify the most relevant topics of the corpus made up of documents such as theses and dissertations of that graduate program, such as the terms that constitute each topic as well as their respective weights. In the topic modeling we set a Latent Dirichlet Allocation model to identify 6, 8, 10, 12, 14, 16, 18 and 20 topics along with the data corpus. This allowed us to scientifically map the documents that we analyzed. The results obtained when the model was set to 14 topics were more cohesive and presented less noise and so allowed us to assume the names of the topics more assertively and to correlate the fields of research of the graduate program of the Federal University of Minas Gerais.