Tweet polarity classification focused on positive and negative term frequency ratio

With the wide spread of SNS (including Twitter, Facebook, and Flickr), there is a great demand for analyzing the associated Web contents consisting of a vast amount of opinions posted from anonymous users. Such opinions usually have explicit or implicit polarities. The polarity determination for short texts like Twitter is, however, very difficult. In this paper, we propose a method for sentiment analysis in Twitter, focusing on the proportion of term frequencies between positive and negative polarities. We compare our proposed method against several other methods. From the experiments using SemEval2016 Twitter data, we demonstrate that our proposed method outperforms the same team in terms of F1-score.