Key points are not available for this paper at this time.
An education system consists of imparting knowledge, assessing the learners, collecting feedback, analysing it, and taking proper measures to improve it. All these measures ultimately improve learner satisfaction which gets reflected in the learner performance and makes the course popular. Due to the corona pandemic, educational institutes and universities have shifted their activities online. In such a scenario, it is challenging to identify a course's popularity and the factors that make it popular. In this work, the authors have performed aspect-based sentiment analysis of learner feedback to identify the essential aspects of creating a popular course. The traditional methods for analysing feedback for sentiment analysis require manual intervention and are tedious. The authors have proposed a framework that identifies learner reviews' aspect level sentiment polarity and selected the significant aspects impacting a course's popularity. Experimental work is conducted over a large-scale real-world education dataset containing around 110 K learner comments (referred to as Educational Dataset now onwards) collected from Coursera1 and learner data from academic institution MSIT2. Different word embeddings viz., FastText, Word2Vec (Word2Vector), GloVe (Global representation of Vectors), and user-created embeddings are used for CNN (Convolution Neural Network) and LSTM (Long Short Term Memory) models. These models are further compared with BERT (Bidirectional Encoder Representation from Transformers), EvoMSA (Evolutionary Multilingual Sentiment Analysis) and ELMo (Embeddings from Language Model) classifiers to find the method that works best for identifying the aspects that impact the popularity of a course. Based on the experimental results over the collected Dataset, BERT based model provides the best results with an accuracy of 94.27%
Jatain et al. (Thu,) studied this question.