Semantic Book Recommender

An AI-powered book recommendation system that uses semantic search and emotion analysis to provide personalized book suggestions based on natural language descriptions.

Features

• Semantic Search: Find books based on natural language descriptions rather than simple keyword matching
• Emotion-Based Filtering: Filter recommendations by emotional tone (Happy, Surprising, Angry, Suspenseful, Sad)
• Category Filtering: Filter books by genre categories (Fiction, Nonfiction, Children's Fiction, Children's Nonfiction)
• Interactive Web Interface: User-friendly Gradio dashboard for exploring recommendations
• Visual Gallery: Display book covers with titles, authors, and descriptions

Architecture

The system consists of several components:

1. Vector Search Engine (vector-search.py): Creates embeddings from book descriptions using OpenAI embeddings and stores them in a Chroma vector database for semantic similarity search
2. Text Classification (text-classification.py): Categorizes books using zero-shot classification with Facebook's BART model
3. Sentiment Analysis (sentiment-analysis.py): Analyzes emotional content of book descriptions using DistilRoBERTa emotion classifier
4. Gradio Dashboard (gradio-dashboard.py): Provides an interactive web interface for searching and browsing recommendations

Installation

1. Clone the repository:

git clone https://github.com/tarek-gritli/semantic-book-recommender.git
cd semantic-book-recommender

2. Create and activate 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. Set up environment variables: Create a .env file in the project root and add your OpenAI API key:

OPENAI_API_KEY=your_api_key_here

Data Requirements

The system expects the following CSV files:

• books.csv: Original book dataset
• books_cleaned.csv: Preprocessed book data with descriptions obtained through the Semantic Book Recommender Kaggle notebook
• books_with_categories.csv: Books with category classifications
• books_with_emotions.csv: Books with emotion analysis scores

Usage

Running the Web Interface

Launch the Gradio dashboard:

python gradio-dashboard.py

The interface will open in your browser where you can:

• Enter a natural language description of the type of book you're looking for
• Select a category filter (optional)
• Select an emotional tone filter (optional)
• View recommended books in a visual gallery

Processing Pipeline

If you need to process new book data:

1. Prepare vector embeddings:

python vector-search.py

2. Classify book categories:

python text-classification.py

3. Analyze emotional content:

python sentiment-analysis.py

Technical Stack

• Language Models:
  • OpenAI Embeddings for semantic search
  • Facebook BART for zero-shot classification
  • DistilRoBERTa for emotion analysis
• Vector Database: Chroma for efficient similarity search
• ML Framework: Transformers, LangChain
• Web Interface: Gradio
• Data Processing: Pandas, NumPy

Project Structure

semantic-book-recommender/
├── gradio-dashboard.py       # Web interface
├── vector-search.py          # Embedding generation and search
├── text-classification.py    # Category classification
├── sentiment-analysis.py     # Emotion analysis
├── requirements.txt          # Python dependencies
├── books*.csv               # Book datasets
├── tagged_descriptions.txt  # Processed descriptions for embedding
└── .env                     # Environment variables (not in repo)

Dependencies

Key dependencies include:

• langchain and langchain-chroma for vector search
• transformers for NLP models
• gradio for web interface
• pandas for data processing
• openai for embeddings

See requirements.txt for complete list.