Multilabel Classification of Student Feedback Data Using BERT and Machine Learning Methods

Studying student feedback is essential for educational institutions to provide good services to their students. The main purpose of evaluation is to improve the services offered to students through the interest monitoring information system. The guardian of each learner should provide feedback on infrastructure and learning services so that the educational institution can improve its services. In addition, the purpose of evaluation is to investigate student inquiries and receive responses from the appropriate departments to ensure the successful delivery of student services. Automatic classification of student feedback is necessary to improve response time and service quality. Student feedback should immediately follow the service to the relevant department, therefore the automated system classifies the feedback according to the unit handling it, prioritizing the most rapid development of the system. Each student feedback can be handled by more than two units, so the problem includes multilabel classification. This study aims at multi-label classification of student feedback data. This study uses a Bidirectional Encoder Representation from Transformers (BERT) to derive word vectors from student feedback data. In this study, several machine learning methods such as Support Vector Machines(SVM), K-Nearest Neighbors(KNN), Random Forests(RF), and Decision Trees(DT) are used to classify multi-label student feedback and compare their performances. This dataset consists of an assessment of the guardianship information system for 3323 students with the composition of the experiment using a comparison of 80% training data and 20% testing data. The SVM method with linear kernel has the best performance as evidenced by the accuracy of 82% and F1 value of 90%.