Evaluation of potential features present in short texts in spanish in order to classify them by polarity

Abstract

This work describes the identification and evaluation process of potential text markers for sentiment analysis. The evaluation of the markers and their use as part of the feature extraction process from plain text that is needed for sentiment analysis is presented. The evaluation of text markers obtained as a result of systematic analysis from a corpus over a second one allowed us to identify that emphasized positive words that tend to appear in positive text posts. The second corpus allowed us to evaluate the relation between the polarity of morphological text markers and the text they appear in. The evaluation of the markers for polarity detection task, in combination with a polarized dictionary, produced polarity classification average precision of 0.56 % using only three markers. These are promising results if we compared them to the top 0.69 % obtained using more features and specialized dictionaries for the same task.

Keywords

• sentiment analysis
• information gain
• feature vectors
• polarity
• classification