Sentiment Analysis of Social Media Comments in Mauritius

Sentiment analysis is becoming increasing important with the rise in the amount of content on social media. However, sentiment analysis remains challenging for under-resourced languages such as Kreol Morisien (KM), the native language of Mauritius. In fact, it has been observed that in Mauritius, social media comments often consist of more than one language among English, French and Kreol Morisien. In this paper, we first create an annotated dataset of 1300 sentences and then outline a framework through which sentiment analysis can be performed on social media comments. We propose a KM sentiment analyzer using two algorithms namely Support Vector Machine (SVM) and Multinomial Naïve Bayes (MNB). Our results show that SVM outperforms MNB for sentiment analysis in Kreol Morisien, achieving an accuracy of 66.15% after pre-processing techniques stopwords removal and spell checking are applied. This paper highlights the need to develop further tools in order to enable natural language processing of Kreol Morisien.