Mineração de dados: um estudo das emoções acerca do Festival Folclórico de Parintins

Abstract

Communications shared by users on social networks carry a large amount of information that can be quantified and analyzed using appropriate statistical procedures. Machine learning models, lexical dictionaries, neural networks, among others, are examples of effective information classifiers. These techniques, besides ensuring reliable results, offer advantages such as saving time and reducing costs, since data are captured almost in real time through web scraping, avoiding the need for researchers to go into the field when using traditional procedures. This study aims to identify and analyze the sentiments and emotions triggered by the Parintins Folklore Festival among supporters, participants, fans, and tourists. To do so, the sentiment analysis technique was used, employing lexical analysis and the Naive Bayes classification model, both tools from machine learning. For the Naive Bayes classification, the dataset was labeled by humans as positive or negative and divided into four considered splits (50% and 50%, 60% and 40%, 66% and 34%, and 70% and 30% for training and testing, respectively). The results from the lexical analysis show a predominance of positive sentiments (67%) in relation to negative ones (33%), with substantial emphasis on the emotions of trust, joy, and surprise. It is worth noting that negative emotions such as sadness, fear, and anger appear quite representative for a festive moment. The Naive Bayes classifier achieved an accuracy of 76%, a median result for a classification model, indicating that out of 10 analyzed texts, it correctly classifies 8.