LLM-Powered Semantic Book Recommender

A user-friendly, interactive book recommendation system that leverages large language models (LLMs), semantic embeddings, and NLP techniques to provide precise content-based recommendations. Users can filter by category and emotional tone, enabling nuanced book discovery beyond keyword matching.

---

Table of Contents

• Project Overview
• Features
• Technologies Used
• How It Works
• Setup & Installation
• Usage
• Challenges & Solutions
• Outcome & Learnings
• Future Improvements
• Credits

---

Project Overview

This project builds a semantic book recommender using modern NLP techniques and LLMs. Book descriptions are transformed into vector embeddings, allowing users to find books that closely align with their interests. The system incorporates:

• Semantic search via vector embeddings
• Zero-shot classification for category normalization
• Sentiment/emotion analysis for mood-based filtering
• Interactive dashboard using Gradio

---

Features

• Semantic Book Search: Find books based on meaning rather than exact keywords.
• Category Filtering: Standardized via zero-shot classification.
• Emotion Filtering: Filter books by emotional tone using sentiment/emotion analysis.
• Interactive Dashboard: Real-time exploration of recommendations with thumbnails, descriptions, and author info.
• User-Friendly Interface: Designed for non-technical users.

---

Technologies Used

• Data Processing: Python, Pandas, NumPy
• LLM & Embeddings: OpenAI embeddings, Hugging Face Transformers
• Vector Database & Retrieval: LangChain + Chroma
• Classification & Sentiment Analysis: Hugging Face Zero-Shot Classifiers, fine-tuned RoBERTa
• Dashboard / Interface: Gradio
• IDE / Tools: PyCharm

Why: Transformers provide contextual embeddings; LangChain + Chroma enable semantic search; zero-shot classification simplifies messy book categories; Gradio creates a practical interactive dashboard.

---

How It Works

1. Data Acquisition & Cleaning:

   • Dataset: 7K Books Kaggle Dataset
   • Cleaning includes handling missing descriptions, standardizing categories, and creating a combined title/subtitle field.

2. Embeddings & Vector Database:

   • Book descriptions are converted into embeddings using Transformer-based models.
   • Embeddings are stored in a Chroma vector database for fast similarity search.

3. Query Handling & Recommendation:

   • User input is embedded and compared against stored book embeddings.
   • Top-k similar books are retrieved.

4. Filtering & Enhancement:

   • Category filtering: via Hugging Face zero-shot classification
   • Emotion filtering: fine-tuned RoBERTa model assigns emotional tones to book descriptions

5. Dashboard Display:

   • Results presented on a Gradio interface with book thumbnails, authors, and filtered recommendations.

---

Setup & Installation

1. Clone the repository:

[git clone https://github.com/Sarach-git/Book-Recommendation.git
cd book-recommender