Mining opinions from instructor evaluation reviews: A deep learning approach

Abstract Student evaluations of teaching (SET) provides potentially essential source of information to achieve educational quality objectives of higher educational institutions. The findings can be utilized as a measure of teaching effectiveness and they may aid the administrative decision‐making process. The purpose of our research is to establish an efficient sentiment classification scheme on instructor evaluation reviews by pursuing the paradigm of deep learning. Deep learning is a recent research direction of machine learning, which seeks to identify a classification scheme with higher predictive performance based on multiple layers of nonlinear information processing. In this study, we present a recurrent neural network (RNN) based model for opinion mining on instructor evaluation reviews. We analyze a corpus containing 154,000 such reviews, with the use of conventional machine learning algorithms, ensemble learning methods, and deep learning architectures. In the empirical analysis, three conventional text representation schemes (namely, term‐presence, term‐frequency [TF], and TF‐inverse document frequency schemes) and four word embedding schemes (namely, word2vec, global vector [GloVe], fastText, and LDA2Vec) have been taken into consideration. The predictive performance of supervised machine learning methods (such as, Naïve Bayes, support vector machines, logistic regression, K‐nearest neighbor, and random forest) and three ensemble learning methods have been examined on word embedding schemes. The extensive empirical analysis indicates that deep learning‐based architectures outperform the conventional machine learning classifiers for the task of sentiment classification on instructor reviews. For the RNN with attention mechanism in conjunction with GloVe word embedding scheme‐based representation a classification accuracy of 98.29% has been obtained.