309-Error

309-Error — BERT-Based Emotion & Sentiment Analysis Web App

A Flask web application that uses a fine-tuned BERT model to classify text into emotion categories, infer overall sentiment, assign an intensity score, and visualize sentiment progression over time.

Overview

This project was built for Megathon 2024 and combines:

• a Flask-based web interface
• a fine-tuned BERT sequence classification model
• emotion-to-sentiment mapping
• sentiment intensity scoring
• dynamic sentiment trend visualization

The app allows a user to enter text, receive an emotion prediction, view overall sentiment, and track how sentiment changes across multiple inputs.

What the app does

For a given text input, the system:

1. tokenizes the text using a BERT tokenizer
2. runs inference using a fine-tuned bert-base-uncased sequence classification model
3. predicts one of the learned emotion labels
4. maps the predicted emotion to:
   • an overall sentiment (positive or negative)
   • an intensity score
5. stores results and plots sentiment intensity across multiple user inputs over time

Model Details

The repository includes a fine-tuned BERT classification model based on:

• Base model: bert-base-uncased
• Architecture: BertForSequenceClassification
• Classification type: single-label classification
• Number of labels: 6

The app also uses a label encoder to convert model outputs back into human-readable emotion labels.

Repository Structure

megathon-24/
├── app/
│   ├── app.py
│   ├── config.json
│   ├── label_encoder.joblib
│   ├── static/
│   └── templates/
├── trained-model/
├── app.zip
└── README.md

Note: Ensure the model path used in app.py matches the actual model folder name in the repository.

Tech Stack

• Backend: Flask
• Modeling: Hugging Face Transformers, PyTorch
• Visualization: Matplotlib
• Serialization: Joblib
• Frontend: HTML, CSS, JavaScript

How to Run Locally

1. Clone the repository

git clone https://github.com/garimamittal13/megathon-24.git
cd megathon-24

2. Create and activate a virtual environment

macOS / Linux

python3 -m venv venv
source venv/bin/activate

Windows

python -m venv venv
venv\Scripts\activate

3. Install dependencies

pip install flask torch transformers joblib matplotlib scikit-learn

4. Fix the model path if needed

Open app/app.py and make sure the model path matches your repository.

For example, if your model folder is:

trained-model/

then update the loading lines accordingly.

Example:

model = BertForSequenceClassification.from_pretrained('./trained-model')
tokenizer = BertTokenizer.from_pretrained('./trained-model')
label_encoder = joblib.load('./trained-model/label_encoder.joblib')

5. Run the Flask app

From inside the app/ directory:

cd app
python app.py

6. Open the app in your browser

By default, Flask runs at:

http://127.0.0.1:5000

App Routes

The app currently includes these routes:

• / — introduction page
• /home — main input page
• /in-depth-analysis — analysis page
• /predict — returns emotion / sentiment prediction
• /plot — returns sentiment-over-time plot

Example Output

For a text input, the app returns:

• Predicted Label — the emotion class predicted by the model
• Overall Sentiment — positive or negative
• Intensity — a derived sentiment intensity score

Why this project matters

This project demonstrates:

• deployment of a transformer-based NLP model in a web app
• integration of machine learning inference with a user-facing interface
• emotion classification and sentiment reasoning
• lightweight analytics through temporal sentiment visualization
• full-stack hackathon prototyping with AI components

Future Improvements

Potential extensions include:

• confidence score display for all classes
• batch analysis of multiple text entries
• persistent user sessions
• improved visualization dashboards
• deployment on Render / Hugging Face Spaces
• support for chat-style sentiment tracking

Repository

GitHub: garimamittal13/megathon-24