Customer Reviews Sentiment Analysis

This project analyzes customer reviews and classifies their sentiment (positive / negative) using multiple approaches:

• VADER (baseline sentiment analysis).
• TF-IDF + Logistic Regression.
• DistilBERT fine-tuned on Yelp reviews.
• A Streamlit app for interactive predictions.

---

📂 Project Structure

.
├── app
│   └── streamlit_app.py         # Streamlit web app for sentiment prediction
│
├── models
│   ├── distilbert-yelp          # Fine-tuned DistilBERT checkpoints
│   │   ├── best
│   │   └── checkpoint-313
│   ├── tfidf_lr_yelp.joblib     # Saved TF-IDF + Logistic Regression model
│   └── tfidf_lr_confusion_m...  # Confusion matrix plot
│
├── notebooks
│   └── eda.py                   # Exploratory Data Analysis
│
├── screenshots                  # Screenshots of results / Streamlit app
│
├── src
│   ├── baseline_vader.py        # VADER baseline sentiment model
│   ├── data.py                  # Data preprocessing and dataset handling
│   ├── evaluate.py              # Evaluation scripts
│   ├── model_tfidf_lr.py        # TF-IDF + Logistic Regression model
│   └── model_transformer.py     # Transformer (DistilBERT) training pipeline
│
├── .gitignore
├── README.md
└── requirements.txt             # Python dependencies

---

⚙️ Installation

1. Clone the repository:

   git clone https://github.com/MaramHamdy/customer-reviews-sentiment.git
   cd customer-reviews-sentiment

2. Create and activate a virtual environment:

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

3. Install dependencies:

   pip install -r requirements.txt

---

🚀 Usage

Train / Evaluate Models

Run scripts from the src/ folder:

python src/data.py
python src/baseline_vader.py
python src/model_tfidf_lr.py
python src/model_transformer.py
python src/evaluate.py

EDA

python notebooks/eda.py

Run Streamlit App

streamlit run app/streamlit_app.py

This will launch a web app in your browser where you can input customer reviews and see sentiment predictions.

---

📊 Models

• Baseline (VADER) – Rule-based sentiment analyzer.
• TF-IDF + Logistic Regression – Classical ML model.
• DistilBERT – Transformer model fine-tuned on Yelp dataset.

---

✅ Evaluate

• Best Model - DistilBERT with the best accuracy that = 0.941

---

📷 Streamlit Web Screenshots

---

📝 License

MIT License