Improving subjectivity detection for Spanish texts using subjectivity word sense disambiguation based on knowledge

摘要

In this paper, we present a Sentence-level Subjectivity Detection method for Spanish using Subjectivity Word Sense Disambiguation (SWSD) based on Knowledge. We use a classic method of Word Sense Disambiguation, using the Spanish WordNet included in Mutlilingual Central Repository 3.0 and the WordNet-Pr as Knowledge base. Because of the alignment between the WordNet and the SentiWordNet, we use this latter as semantic resource annotated with polarity values to determine when a word expresses subjectivity and objectivity, defining subjectivity levels using a fuzzy clustering algorithm previously. Due to the few resources focused on Sentiment Analysis for Spanish, the Semcor corpus was used for analyzing the attributes to be used. Finally, a Rule-based classifier was created to detect subjective sentences. This method was executed over a Spanish corpus, created in this work. The results show that our approach contributes positively to Subjectivity Detection task, despite of using resources created for English.