Sentiment-Aware Recommendation System

This project implements a sentiment-aware product recommendation system for Amazon Electronics Reviews. It combines collaborative filtering (matrix factorization), sentiment analysis (DistilBERT), key phrase extraction (statistical language model), and explanation generation (instruction-tuned GPT-2) to provide personalized product recommendations with user-friendly explanations. The system is deployed as an interactive Streamlit app, allowing users to input a reviewerID and receive top-5 product recommendations with sentiment-based explanations.

Setup Instructions

1. Clone the Repository:

   git clone https://github.com/lijoraju/sentiment-aware-recommendation.git
   cd sentiment-aware-recommendation

2. Set Up Google Colab:

   • Open Colab, mount Google Drive:

     from google.colab import drive
     drive.mount('/content/drive')

   • Copy the repository to /content/drive/MyDrive/sentiment-aware-recommendation.

3. Install Dependencies:

   !pip install streamlit==1.38.0 pandas==2.2.2 torch==2.4.0 transformers==4.44.2 numpy==1.26.4 scikit-learn==1.5.1 nltk==3.8.1 pyngrok

4. Prepare Data and Models:

   • Place processed_reviews.csv, bert_sentiment_model/, gpt2_explanation_model/, mf_model.pth, key_phrases.txt, sentiment_cache.csv, and explanation_data.csv in /content/drive/MyDrive/sentiment-aware-recommendation.
   • Note: These files are excluded from the repository due to size (see .gitignore).

5. Run the Streamlit App:

   !streamlit run /content/drive/MyDrive/sentiment-aware-recommendation/app.py &>/dev/null&
   from pyngrok import ngrok
   !ngrok authtoken YOUR_NGROK_AUTH_TOKEN
   public_url = ngrok.connect(8501)
   print(f"Streamlit app running at: {public_url}")

Dependencies

• Python 3.11
• streamlit==1.38.0
• pandas==2.2.2
• torch==2.4.0
• transformers==4.44.2
• numpy==1.26.4
• scikit-learn==1.5.1
• nltk==3.8.1
• pyngrok

Usage

1. Open the Streamlit app via the ngrok URL.
2. Select a reviewerID from the dropdown (loaded from processed_reviews.csv).
3. View top-5 recommendations, including:
   • Product ASIN (e.g., B000067O7T).
   • Sentiment (Positive/Neutral).
   • Key phrases (e.g., “great camera, sturdy case”).
   • Explanation (e.g., “Product B000067O7T is recommended because...”).

Example Output

Recommendation 4: Product B000067O7T
Sentiment: Positive
Key Phrases: great camera, sturdy case, case perfect
Explanation: Product B000067O7T is recommended because users found it to have a great camera, sturdy case, and perfect case, making it an excellent choice for photography enthusiasts.

Evaluation

Run evaluate.py to compute:

• RMSE for matrix factorization recommendations.
• Precision and recall for sentiment classification.

python evaluate.py

Contact

For questions or contributions, contact lijoraju.