Avaliação de métodos de construção de redes e detecção de comunidades no agrupamento de textos

Abstract

Due to the large amount of data produced daily in text format, whether publicly on social networks or privately within companies, there is a need to analyze and extract information from them. The goal is to turn them into useful tools, such as translation systems and virtual assistants. The area of Natural Language Processing, in conjunction with Machine Learning, provides the necessary technologies for such an objective. One of the most explored tasks in this context is the clustering of documents through unsupervised classification. Document clusters can provide a description of the subjects covered by a collection of documents, representing, in general, categories or themes. Considering this task, in addition to the traditional clustering algorithms, such as k-Means, approaches based on networks have been gaining notoriety in the literature, which build a network from the document collection and use community detection to find groups of documents representing similar themes. These approaches initially need the construction of a network from the documents analyzed, and several algorithms can be used for this purpose, which produces networks with distinct topological characteristics, directly interfering with the quality of the cluster. In this context, the aim of this study is to analyze the influence of network construction algorithms in the clustering of texts. It seeks to assess whether the different ways of building networks can influence the generation of community structures that are representative considering the classes of text documents.