Análise exploratória de tweets utilizando modelagem de tópicos para textos curtos: caso Olimpíadas Rio 2016

Abstract

Topic modeling is a data mining problem that aims to extract, given a document collection, the main topics that represent the subjects covered in the collection documents. A document, which can be short or long, can be defined as a mixture of topics, being a set of words ordered by their probability of occurrence. Conventional approaches for topic modeling such as LDA and PSLA have been used in long documents and when used it in short texts may not work well since conventional topic models suffer from the severe data sparsity. So, in this paper, an exploratory analysis is performed in the Twitter database applying the Biterm Topic Model (BTM) to discover the main topics discussed during the Rio 2016 Olympic Games. As a result, the topics obtained represents the collection of documents and reflects the events that occurred at the Olympic Games, especially those that made reference to Brazil.