This project applies a deep learning–based natural language processing approach to analyze Japanese student feedback collected via Google Forms from project-based English classes, using a pretrained Japanese BERT model. The analysis focuses on identifying students' favorite and least favorite projects, reasons for their preferences, and suggestions for improvement. The goal is to extract actionable insights to enhance curriculum design and student engagement.
An earlier exploratory version of this project (2023) focused on morphological analysis and keyword extraction. In 2025, the project was reviewed and redesigned to focus specifically on a Japanese Sentence-BERT model to improvement responses to identify latent semantic themes via unsupervised clustering (to identify meaning-based themes in student feedback).
| Cluster | Semantic Theme | Description |
|---|---|---|
| Cluster 1 | Practice Time & Preparation | Requests for longer rehearsal periods, increased practice time, and more opportunities to prepare or collaborate before presentations. |
| Cluster 0 | Task Structure & Instruction Clarity | Feedback indicating unclear instructions, overly broad task scopes, or a need for clearer examples, constraints, or scaffolding. |
| Cluster -1 | Miscellaneous / Individual Feedback | Diverse, low-frequency suggestions that did not form a stable semantic cluster, including scheduling concerns and project-specific difficulties. |
| Favorite Project | Cluster -1 (Misc.) |
Cluster 0 (Structure) |
Cluster 1 (Practice) |
|---|---|---|---|
| Demonstration Dialogue | 4 | 0 | 0 |
| Country Presentation | 36 | 11 | 3 |
| Skit Festival | 37 | 8 | 5 |
| Interview Speech | 4 | 7 | 1 |
Data Privacy Notice
All student responses were anonymized and analyzed in aggregate.
Data collection followed institutional consent and privacy guidelines.