Comparison of Clustering Techniques in Text Documents in Portuguese

Managing the vast amount of text data in the digital world is a complex challenge. An effective approach to tackle it is through the technique of text document clustering. This study evaluated the performance of three clustering algorithms — K-Means, Single Linkage, and Gaussian Mixture Model (GMM) — in clustering Brazilian Portuguese news articles using BERTimBau, a Portuguese variant of the BERT model, for preprocessing. Metrics such as accuracy, F1-score, Rand index, and Jaccard coefficient were used for evaluation. The results of these metrics indicated that Single Linkage achieved the best overall performance, surpassing K-Means and GMM in most of the evaluated criteria.