🔍 Subtheme Sentiment Analysis using BERT

This project focuses on performing multi-label subtheme sentiment analysis using BERT (Bidirectional Encoder Representations from Transformers). Each input text (e.g., a customer review) is analyzed to detect multiple subthemes (like garage service, ease of booking, value for money, etc.) and their corresponding sentiments (positive/negative).

---

📌 Project Goals

• Classify multiple subtheme-sentiment labels per review
• Fine-tune bert-base-uncased for multi-label text classification
• Prepare the model for deployment or reuse in other projects

---

🧪 Dataset

• The input is a structured CSV file where:
  • Column text contains customer reviews
  • Other columns are binary labels (0/1) representing sentiment for each subtheme.

Example:

text	garage_service_positive	ease_of_booking_positive	value_for_money_positive
"Tyres delivered quickly, smooth experience"	1	1	0

---

⚙️ Model Used

• bert-base-uncased from HuggingFace
• Fine-tuned for multi_label_classification
• Uses sigmoid activation and binary cross-entropy loss

---

🚀 Quickstart

✅ 1. Clone Repository

git clone https://github.com/Aswin-Cheerngodan/Subtheme-Sentiment-Analysis.git
cd subtheme-sentiment-analysis

✅ 2. Install Dependencies

python -m venv myenv
myenv\Scripts\activate
pip install -r requirements.txt

✅ 3. run streamlit app

streamlit run src/main.py

🌐 Access the App

subtheme-sentiment-analyzer: http://localhost:8501

🧪 Example Usage

• Type a customer review related to the garage.
• It finds the subthemes along withe their sentiment.
• Optionally you can view scores of all subthemes.

📬 Contact

Questions or contributions? Open an issue or reach out at aachu8966@gmail.com