2019 Volume 41 Issue 1 Pages 125-137
In this study, we used topic modeling based on latent Dirichlet allocation with about 60,000 datasets collected over nine years at a university to analyze the free descriptions obtained through course evaluation questionnaires, which have been considered to be difficult to utilize because of various limitations including their manual classif ication. We labeled the 170 extracted topics according to a series of steps and verif ied the validity of the labels. The results conf irmed that the labels had an adequate level of validity, demonstrating that classif ication via topic modeling is suitable for human senses from a holistic viewpoint. In this analysis, we also linked free descriptions with information on courses and visualized information using cross-tabulation to see the proportion of topics included in each of the courses and characteristics of the topic distribution in each course. This analysis method can be applied in the f ields of institutional research and learning analysis in future studies.
