Main Article Content
Sentiment Analysis has been an interesting and popular research area encouraging researchers and practitioners to adopt this tool in various fields such as the government, health care and education. In education, instruction evaluation is one of the activities that sentiment analysis has served. Though, it is a common practice that educational institutions periodically evaluate their teachers’ performance, students’ comments which are rich in insights are not easily taken into account because of lack of automated text analytics methods. In this study, supervised machine learning algorithms were used. Experiments were conducted to evaluate base models employing naïve bayes, support vector machines and logistic regression in comparison to ensemble combining the three. Random forest, an ensemble learning algorithm was also experimented. Machine learning techniques such as term-frequency – inverse document frequency (TF-IDF) and ngram were also explored to devise a model with the highest possible accuracy. Results show that in this case, tf-idf vectorization does not show significant improvement in sentiment classification. On the other hand, ngram vectorization improve performance of base models and has potential to improve ensemble models. Random forest showed higher performance measures than base models and ensemble of the three base models. However, it did not outperform ngram combined with support vector machines. In future work the model with highest accuracy found can be embedded in a sentiment analysis tool for students’ feedback on teaching performance. More advanced transformation techniques and other ensemble techniques may be explored to further improve accuracy in sentiment classification.