Comparing Sentiment Analysis Models to Classify Attitudes of Political Comments on Facebook (November 2016)

This paper is a preliminary study which compares nine ML methods of sentiment analysis aimed towards classifying a corpus of 5.3 million messages of the public on Facebook pages of incumbent politicians. Two sentiments were examined: the general attitude of a comment and the attitude of the comment towards the content of a political post. Our results show that Logistic Regression outperformed the other eight ML models in terms of accuracy and F-measures. Also, we found that n-gram representation performed best. An interesting finding is a difference in success rate when classifying attitude in general vs. attitude towards the content in the political context.