Sentiment & Insight Extractor

Turn raw feedback into instant insights.

This Streamlit app analyzes customer reviews, survey responses, or any text dataset to:

• Classify sentiment (positive / negative / neutral)
• Extract top recurring themes using KeyBERT
• Export detailed results to CSV

Built 100% with free, local tools, no paid APIs required.

Features

• 📂 Upload a CSV of feedback (must include a text column)
• 🧠 Transformer-based sentiment analysis (DistilBERT)
• 🔑 Theme extraction with KeyBERT + MiniLM embeddings
• 📊 Interactive charts (sentiment distribution + top themes)
• ⬇️ One-click download of results

Tech Stack

• Python
• Streamlit — web UI
• Hugging Face Transformers — sentiment classification
• KeyBERT — keyword & theme extraction
• SentenceTransformers — MiniLM embeddings
• pandas — data handling

Project Structure

sentiment-insights/
├─ app.py # Main Streamlit app
├─ requirements.txt # Dependencies
├─ sample_data/
│ └─ reviews.csv # Example dataset
└─ README.md

Getting Started

1. Clone the repository

   git clone https://github.com/<your-username>/sentiment-insights.git
   cd sentiment-insights

2. Create a virtual environment

   python -m venv .venv
   source .venv/bin/activate   # On Windows: .venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

4. Run the app

   python -m streamlit run app.py

Tips

• For multilingual text, select the NLPtown model in the sidebar; for mixed languages, switch KeyBERT’s embedding to the multilingual MiniLM.

• If you get fewer themes than requested, reduce Min docs per theme to 1 and use n-grams (1,2).

• On macOS, if PyTorch install fails, try:

  python -m pip install torch --index-url https://download.pytorch.org/whl/cpu

Future Improvements

• Abstractive summarization of top themes

• Support for scanned PDFs with OCR

• Precision/recall evaluation for theme extraction

• Integration with multilingual sentiment models