How pavithras2022b-bit/multilingual-emotion-detection works
This repository’s runnable product is a Streamlit interface for classifying Tamil or Telugu review text into an emotion label and displaying confidence scores. The app loads a fine-tuned transformer and label encoder, tokenizes submitted text, runs inference, and renders the result. The README also describes a broader collection, preprocessing, labeling, training, and evaluation workflow, but its notebook was not inspected and the referenced transformer directory is absent from the file tree; those documented stages are shown as documentation-only context.
Main entry points: app.py Emotion analyzer · README.md Text preprocessing · mbert_label_encoder.pkl Label encoder
Components 13
Review application
- Emotion analyzer
app.py - Transformer classifier
app.py - Confidence results
app.py
Data and models
- Label encoder
mbert_label_encoder.pkl - Fine-tuned weights
- Balanced reviews
balanced_dataset_light.csv - Preprocessed reviews
preprocessed_dataset.csv
Documented workflow
- Review collection
README.md - Text preprocessing
README.md - Zero-shot labeling
README.md - Model training
README.md - Model evaluation
README.md
Other components
- Review submitter
Connections 9
- Review submitter → Emotion analyzer: enters review
app.py - Emotion analyzer → Transformer classifier: runs inference
app.py - Emotion analyzer → Label encoder: loads labels
app.py - Emotion analyzer → Fine-tuned weights: loads model
app.py - Emotion analyzer → Confidence results: renders scores
app.py - Review collection → Text preprocessing: feeds reviews
README.md - Text preprocessing → Zero-shot labeling: prepares text
README.md - Zero-shot labeling → Model training: labels data
README.md - Model training → Model evaluation: produces models
README.md