Construção automática de um dicionário emocional para o português

Abstract

It has been verified in the last few years, the increasing of data contents and easy access to Web allowing an excessive gathering of text from several natures. Despite of all positive aspects and the promising potential, appears a new problem related to the need of develop tools and methods capable to treat those contents in opinions and feelings. The evaluation of Web contents is not an easy task. The content evaluation techniques belong to feelings analysis area and there is a lot of proposed works. This thesis follows a different course by evaluating the web contents to European Portuguese language. The initial criterion adapted is to build the bases to create feeling classifiers. The emotional lexicon is the base to the most of the feeling analysis methods. Regardless of being a lot of available resources to the scientific community, has been verified after big research, that there is no similar resource to European Portuguese language. With the interest of most companies and individuals to gain product information from data web in real time, there is a need to build an emotional Portuguese lexicon that might be use to feeling analysis. To deal with this missing, an emotional Portuguese lexicon has been built. The feeling analysis methods use lexicons, built manually or semi automatic, appearing the problem of increasing the linguistic lexicon acknowledgment associated to the way of how they are built. Considering the feelings has the key of the process, it’s important to know that feelings are synonymous of subjectivity. The goal of this dissertation is to build an automatic European Portuguese subject lexicon using statistical techniques. The lexicon construction base is the corpora which will be used. To do this research is required a subjective corpus (built by blog text) and an objective corpus (built by CETEMP´ublico journalistic corpus text). The corpora have been chosen based in Pais [21] research in the similarity demonstration between a blog amount and a journalistic corpus, relatively to a corpus built by subjective and objective texts. To identify the text subjectivity has been used the information from the morphologic categories (part-of-speech) of simple and compound (n-grams) words. These subjectivity evidences were extracted with tools that work with the automatic process. This work demonstrate that it is possible to build a subjectivity lexicon to the European Portuguese language applying statistical techniques and using manually not noted corpora as well as subjectivity evidences automatic extraction tools.