LLM-based Recommendation Systems enhanced by RAG

Overview

This project is a Retrieval-Augmented Generation (RAG) based movie recommendation system built using GPT-4o mini, DeepSeek-V3, FAISS, and LangChain. It retrieves movies from the MovieLens dataset and generates structured, contextualized recommendations.

Features

• Semantic Retrieval: Uses FAISS to find similar movies based on user queries.
• LLM-based Recommendation: Generates movie suggestions and explanations using GPT-4o and DeepSeek-V3.
• Hybrid Filtering: Combines metadata filtering, embedding-based retrieval, and ranking to refine recommendations.
• API Integration: Exposes endpoints for external usage via Flask or FastAPI.

Installation

# Clone the repository
git clone https://github.com/ZenanJ/ECE1508.git
cd ECE1508

# Install dependencies
pip install -r requirements.txt

Configuration

Modify src/utils/config.py to update hyperparameters, API keys, and retrieval settings.

Create .env file based on the .env.sample file content.

cp .env.sample .env

And input your API secret key in the .env file.

Usage

Run in Gradio UI mode

python src/app.py --mode ui --model remote --llm openai

The chatbot interface will be available at http://localhost:7860.

Run in command-line mode

python src/app.py --mode cli --model remote --llm openai

Repository Structure

ECE1508/
│── evaluation/                 # evaluation method and test data
│── images/                     # example images
│── movielens_data/             # Directory for dataset storage
│   ├── raw/                    # Raw MovieLens dataset files
│   ├── processed/              # Preprocessed data and embeddings
│── src/                        # Main source code
│   ├── retrieval/              # Retrieval system (FAISS, embeddings)
│   │   ├── index_builder.py    # Build FAISS index
│   │   ├── retriever.py        # Retrieve similar movies from index
│   ├── generation/             # LLM-based response generation
│   │   ├── prompt_templates.py # Define structured prompts for GPT-4o & DeepSeek-V3
│   │   ├── generator.py        # Generate movie recommendations using LLM
│   ├── utils/                  
│   │   ├── config.py           # configuration file where load API key from .env file
│   ├── app.py                  # Using Gradio build a chatbot interface
│── .env.sample                 # sample format for .env file
│── requirements.txt            # Required dependencies
│── README.md                   # Project documentation

Examples

Gradio Chatbot Interface

Retrieved Results Directly from FAISS vectorDB