Amazon Recommender System with Graph Neural Networks (GNNs)

Welcome to the Amazon Recommender System project! This repository implements a recommendation system for Amazon products using Graph Neural Networks (GNNs). The system goes beyond traditional recommendation algorithms like collaborative filtering by leveraging advanced graph-based methods to uncover deeper insights into user-product interactions.

---

📌 Overview

Motivation

Recommendation systems play a vital role in personalizing user experiences. Amazon's ability to recommend products effectively relies on advanced algorithms. This project explores Graph Neural Networks to simulate such a system.

Highlights

• Built using the Amazon 2018 dataset with a subset processed using the 5-kcore method.
• Utilizes GraphSAGE and Graph Attention Networks (GAT) for node embeddings and aggregation.
• Benchmarked against traditional models like KNN and SVD.

---

🛠️ Features

1. Graph Representation:

   • Nodes: Users and products.
   • Edges:
     • Reviews: Connect users to products with ratings.
     • Also Bought: Connect products frequently purchased together.

2. Node Embeddings:

   • Product Nodes: Generated using sentence-transformer/all-MiniLM-L6-v2 for descriptions and one-hot encoded categories.
   • User Nodes: Initialized as blank and updated dynamically based on interactions.

3. GNN Models:

   • GraphSAGE: Learns inductive representations of nodes.
   • GAT: Assigns attention weights to neighbors for better aggregation.

4. Explainability:

   • Integrated gradients via Captum to interpret model predictions.

5. Evaluation:

   • Metric: Root Mean Squared Error (RMSE).
   • Comparison against traditional methods and a state-of-the-art Graph-based Hybrid Recommendation System (GHRS).

---

🚀 Getting Started

Prerequisites

• Python 3.9
• PyTorch Geometric
• Hugging Face Transformers
• Other Python libraries (see requirements.txt)

Installation

1. Clone the repository:

   git clone https://github.com/aiden200/ARS.git
   cd ARS
   

2. Install the required Packages: pip install -r requirements.txt.
3. Run the Cells in ARS.ipynb

The data is already downloaded in the repository, so there is no need to uncomment the data collection line within the notebook.

Our paper: Final_paper