Identificação automática da presença social em discussões online escritas em português

Abstract

This M.Sc. dissertation presents a method that allows the automatic identification of messages from distance learning online forums written in Brazilian Portuguese. In particular, it analyzes the problem of coding discussion messages according to the categories of social presence, an important construct of the Community of Inquiry (CoI) model widely used in online learning. Although there are coding techniques for social presence in the English language, the literature is still lacking in methods for other languages, such as Portuguese. The method proposed here uses a set of characteristics derived from the frequency of words and 158 characteristics extracted from two resources, LIWC and Coh-Metrix, available for textual analysis using Text Mining techniques, to create a classifier for each one of the three categories of social presence (Affective, Interactive and Cohesive). For that, three types of algorithms were used, Random Forest, AdaBoost and XGBoost where the best model developed used the XGBoost algorithm reaching 85.68% accuracy and Kappa index (k) of 0.71, which represents a substantial agreement, and is well above the level of pure chance. This work also provides an analysis of the nature of social presence, observing the classification characteristics that were most relevant to distinguish the three categories of presence and a comparative analysis on the main characteristics identified in the phases of social presence in different domains.