Análisis de opiniones en español

Abstract

The present thesis is about the Natural Language Processing task known as Sentiment Analysis. The thesis exposes a deep study concerning the automatically treatment of opinions written in Spanish. The processing of opinions can be tackled at three levels, of which the document level was the selected one to carry out the different experiments. In the context of supervised classification, an analysis of different methods to measure the relevance of features will be described. Furthermore, an unsupervised algorithm based on the extension of the meaning of words will be exposed. Moreover, the good results reached by means the combination of dissimilar classifiers will be described. Also, two corpus of opinions and a list of opinion bearing words will be described. Finally, we have to remark that all the assessments will be done on texts from microblogging networks, and opinions from reviews web sites.