Sentiment Analysis and Semantic Network Analysis of University Lecture Evaluation Narrative Questionnaire through A University Practice

Abstract

Objectives The purpose of this study was to derive directions and implications for quality of education through sentiment analysis and semantic network analysis of university descriptive course evaluations. Methods The analysis data consisted of 243,900 descriptive lecture evaluations from the first semester of 2020 to the first semester of 2023 at University A. Emotion words were classified in the order of positive, negative, and neutral, and the frequency and ratio of descriptive lecture evaluation contents by year, subject, and college were calculated. N-gram and clustering were analyzed using Textome, and network centrality between keywords was analyzed and visualized using Ucinet 6. Results First, in the sentiment analysis trend for descriptive lecture evaluation questions by year, course com-pleted, and university, positive opinions were found to be dominant. Second, as a result of the keyword frequency and N-gram analysis of the lecture evaluation contents, the frequency of ‘class’ and ‘thank you’ was high in pos-itive evaluations, and the frequency of ‘class’, ‘assignment’, and ‘unsatisfactory’ was high in negative descriptive course evaluations. In addition, ‘class’ was expanded around the keywords ‘content’, ‘progress’, ‘time’, and ‘regrettable’. Third, the results of semantic network analysis between keywords in the lecture evaluation contents showed that in the positive type, ‘gratitude’, ‘class’, ‘hard work’, ‘explanation’, ‘student’, ‘kindness’, ‘help’, ‘understanding’, ‘assignment’, ‘time’ were the most important and connected keywords. In the negative course evaluation, keywords including ‘class’, ‘lecture’, ‘professor’, ‘assignment’, ‘disappointing’, ‘difficult’, ‘anxiety of test’ were analyzed as the most connected. Conclusions This study showed the possibility of suggesting directions for education improvement through de-scriptive lecture evaluation using sentiment analysis and semantic network analysis.