Leveraging Deep Learning for Classifying Learner-Generated Course Evaluation Texts

Abstract

With the growing popularity of massive open online courses (MOOCs), there are many chances to analyze student-generated assessments of course content to learn more about the experiences of learners. Unstructured textual data is typically subjected to manual analysis for qualitative evaluation, which produces a limited knowledge of learners’ experiences. This study looked into the use of a convolutional neural networks-bidirectional long short-term memory network (CNN-BiLSTM) hybrid neural network model to automatically classify significant content in student-written course evaluation documents. Nine categories: “Platforms and tools”, “Overall evaluation”, “Course introduction”, “Course quality”, “Learning resources”, “Instructor”, “Learner”, “Relationship”, “Process”, and “Assessment” were successfully recognized by the artificial neural network-based model. Using a well-established coding framework, an annotated dataset of learner-generated course evaluations was built, and 8,588 MOOC review words from Class Central were analyzed to assess the model’s performance. The CNN-BiLSTM model performed better than the other models when compared to the baseline techniques. It obtained an overall F1 score of 0.7563, an accuracy score of 0.807, a recall score of 0.7552, and a precision score of 0.7612. The CNN-BiLSTM model quickly and efficiently extracts semantic information from contexts by comprehending the local and global features of learner-generated course evaluation texts. With this knowledge, learner-generated course assessments can be managed more effectively, which could improve communication between teachers and students.