Retrieval of Weighted Lexicons Based on Supervised Learning Method

Lexicons are a lexical resource that has been used successfully in sentiment analysis and other areas of natural language processing. Although there are several unweighted lexicons and weighted lexicons, they all achieve poor performance in many applications. This is because they are created in general contexts, and adding more terms to an existing lexicon is complicated. Furthermore, current methods for generating weighted lexicons are complex and not very intuitive. In this article, we show the results of a method to generate weighted lexicons from a tagged corpus. The terms that make up the lexicon, as well as the corresponding weights, are obtained by means of a distance measure that is closely related to the probability that a document belongs to its label. The preliminary results obtained with a corpus of 405 documents show that the method reaches an accuracy of 92.3%.