Automatic Analysis of Facebook Posts and Comments Written in Brazilian Portuguese

Abstract

Social networks and media are becoming increasingly important sources for knowing people's opinions and sentiments on a wide variety of topics. The huge number of messages published daily in these media makes it impractical to analyze them without the help of natural language processing systems.This article presents an approach to cluster texts by similarity and identifying the sentiments expressed by comments on then (positive, negative and neutral, among others) in an integrated manner. Unlike most of the available studies that focus on the English language and use Twitter as a data source, we treat Brazilian Portuguese posts and comments published on Facebook. The proposed approach employs an unsupervised learning algorithm to group posts and a supervised algorithm to identify the sentiments expressed in comments to posts. In an experimental evaluation, a system that implements the proposed approach showed similar accuracy to that of human evaluators in the tasks of clustering and sentiment analysis, but performed the tasks in much less time.