Validação e construção de um dicionário léxico para auxiliar a análise de sentimentos em repositórios de projetos de software

Abstract

Sentiment analysis makes inference about polarities in words that can represent possible emotions. The assertiveness of this classification is important for the results reliability. For this reason, this article investigates, validates and builds the lexicon dictionary, in the context of Software Engineering, using words, emoticons and idiomatic expressions from the SentiStrength-SE tool. An experiment online with 559 questions answered from 48 participants in the Computing area was performed to validate lexical terms agreement from the dictionary. At the end of the data collection, the terms were gathered for validation using a Stack Overflow database to find the results on accuracy, precision, recall and F 1 -score of the new dictionary. The new lexical dictionary has 79% Accuracy and Precision, with 78% Recall and f1-score with a smaller polarity interval than the original dictionary.