Sentiment Classification of Sinhala Content in Social Media: A Comparison between Stemmers and N-gram Features

Abstract

Sentiment classification for non-English languages has gained significant attention from researchers in the past few years with the increasing use of non-English scripts and Romanized scripts for expressing sentiments over social media. In this study, we begin by classifying Sinhala sentiments on social media into positive and negative polarity classes using N-gram feature extraction. N-grams are a contiguous sequence of words or characters of a text. Then we focus on improving the classification accuracy by employing different stemming methods. Stemming is generally used to reduce the dimensionality of the feature set - something which needs to be carried out with great care as over reducing feature dimensionality causes the classification accuracy to decrease. Finally, we compare the accuracy and efficiency of N-gram feature extraction and stemming based sentiment analysis models.