ASPECT-BASED SENTIMENT ANALYSIS OF OPEN-ENDED RESPONSES IN STUDENT COURSE EVALUATION SURVEYS

Abstract

Student surveys are tools used to gather feedback and opinions from students regarding various aspects of their educational experience such as satisfaction with courses, instructors, supervision, facilities, services, and overall campus environment. These surveys play a crucial role to assess and improve the quali- ty of education and overall student experience, and provide insights that can be used to decision makers. There are two primary categories of survey questions: open-ended and closed-ended. Open-ended questions allow respondents to pro- vide detailed answers in their own words, while closed-ended questions offer only predetermined options. For analyzing student surveys, it is common practice to focus on closed-ended questions, while often open-ended responses provide valu- able insights. In this paper, we use sentiment analysis to extract students’ emotions towards various aspects of their educational experience using transformer-based pre-trained language models. The results show that the multilingual XLM-RoBER- Ta demonstrated encouraging results compared to the multilingual BERT model with an accuracy of 0.81 and F1 of 0.80 for the classification of aspects while for the classification of sentiments, obtained 0.94 and 0.92 for accuracy and F1, respectively.