An analysis of educational satisfaction using sentiment analysis: correlation analysis between Likert-scale evaluation and descriptive feedback

초록

This study examines the correlation between multiple-choice evaluations and narrative feedback in corporate training. We analyse satisfaction surveys from 20,160 employees who completed programmes at L Corporation's Training Institute (South Korea) in 2021. Likert-scale ratings are compared with sentiment scores derived from open-ended comments using a Korean-language model hosted on the Hugging Face platform. Likert responses are min-max normalised to the sentiment scale. Results show a statistically significant positive correlation between the two measures (r = .2146, p < .001). Overall, 81% of responses display directional agreement. Subgroup analyses reveal heterogeneity: employees with longer tenure or higher rank tend to express more negative sentiment in free text despite relatively high Likert ratings, and male respondents are more likely than females to voice dissatisfaction. Findings indicate that sentiment analysis yields a complementary, correlation-consistent indicator for HRD programme evaluation, capturing affective and experiential dimensions underrepresented by closed-ended items. Embedding natural language processing into routine evaluation enables scalable analysis of unstructured narratives and supports correlation-based triangulation with conventional metrics. Methodologically, we delineate a robust reproducible pipeline integrating normalised Likert data with transformer-based sentiment scoring. Practically, the approach helps identify aligned and misaligned cohorts for follow-up and targeted improvement.

키워드