SASys: uma arquitetura de Análise de Sentimento usando abordagem lexical

Abstract

Sentiment Analysis or Opinion Mining is a subarea of text mining that aims to detect the opinions, feelings or emotions expressed in a text. It is a multidisciplinary area that ranges from simple Natural Language Processing to sophisticated Machine Learning algorithms. It needs training bases or lexicons that cover peculiarities of the context in which it is applied. Thus, this work proposes to contribute in the field of Opinion Mining presenting a Sentiment Analysis architecture that uses FrameNet as part of its lexical approach, with the objective of discovering the emotional state of the author of the text, through the use of different types of data. This dissertation also presents a polarized network of frames developed from the traditional framework of FrameNet, aiming to detect the polarity of the words through the identification of the evoked frames. A case study with real data was conducted with the support of experts and the results evidence the employability of FrameNet as a lexical approach to sentiment analysis.