Movie Recommendation System with Graph Neural Networks (GNN)

This repository contains a PyTorch implementation of a movie recommendation system using Graph Neural Networks (GNN). The system models user-movie interactions as a bipartite graph, where users and movies are nodes, and edges represent user ratings for movies. By leveraging GNN layers, the model learns to predict user ratings for movies.

---

Requirements

To run this project, install the following dependencies:

• Python 3.8+
• PySpark
• PyTorch
• PyTorch Geometric

Libraries Used

PySpark (Big Data Processing)

• pyspark.sql.SparkSession: Used to create a Spark session.
• pyspark.sql.functions: Includes utility functions for data manipulation, such as:
  • col: Specify a column in a DataFrame.
  • lit: Add constant values to a column.
  • isnan: Check for NaN values.
  • when: Create conditional expressions.
  • count: Count the number of elements in a column.

PyTorch (Deep Learning)

• torch: Core PyTorch library for tensor operations.
• torch_geometric.data.Data: Stores graph data structures.
• torch_geometric.nn.GCNConv: Implements Graph Convolutional Network (GCN) layers.

PyTorch Neural Network Modules

• torch.nn.Linear: Fully connected layer.
• torch.nn.ReLU: ReLU activation function.
• torch.nn.Dropout: Applies dropout regularization to prevent overfitting.
• torch.nn.BatchNorm1d: Batch normalization for stable and faster training.

Dataset

This implementation expects a dataset in the form of a DataFrame with the following columns:

• userId: Unique ID of the user.
• movieId: Unique ID of the movie.
• rating: Rating given by the user to the movie.

Example structure:

userId	movieId	rating
1	101	4.5
2	102	3.0

---

How It Works

1. Data Preprocessing:

   • Convert userId and movieId to unique indices.
   • Create an edge list representing user-movie interactions.
   • Normalize ratings to use as edge attributes.

2. Model Architecture:

   • Three GCNConv layers with batch normalization and ReLU activation.
   • Fully connected layers for regression to predict user ratings.

3. Training:

   • The model is trained to minimize the Huber Loss.
   • RMSE and MAE are used as evaluation metrics.
   • Early stopping is implemented to prevent overfitting.

Results

• Metrics: The model evaluates performance using RMSE and MAE.
• Early Stopping: Stops training if no improvement in RMSE is observed for 20 consecutive epochs.

---

Acknowledgments

This implementation is inspired by PyTorch Geometric and utilizes graph-based learning techniques for recommendation systems. Special thanks to the open-source community for their valuable resources.