Japanese Student Feedback Analysis

Python • BERT • Japanese NLP

Overview

Asked Japanese students to write what they liked, disliked, and wanted improved in project-based English classes.

Cleaned responses and analyzed comments using Japanese BERT embeddings and clustering to find recurring themes.

Used results to simplify curriculum design and remove one weak project.

Results

Source

Student survey responses

Method

Japanese BERT + theme clustering

Main Findings

Students valued creative and cultural projects most

Outcome

One low-value project was removed

Charts

Feedback data Positive themes Improvement themes

BERT Clustering of Student Improvement Feedback

Japanese free-text responses were converted into semantic embeddings and grouped into recurring themes using unsupervised clustering. This helped identify which class projects created the most friction and why.

Cluster 1

Practice Time & Preparation

Students requested more rehearsal time, planning time, and collaboration time.

Cluster 2

Instructions & Structure

Students wanted clearer examples, clearer tasks, and better expectations.

Cluster 3

Mixed / Minor Feedback

Low-frequency comments specific to individual projects.

Where Friction Appeared Most

  • Interview Speech: Higher need for clearer structure
  • Skit Festival: More requests for preparation time
  • Country Presentation: Mostly positive with minor improvements
  • Demonstration Dialogue: Low engagement and weak preference scores

These findings supported removal of one lower-value project and simplification of future curriculum design.

Code

View GitHub Repository