Collaborative Filtering Methods for Paper Recommendation Systems

Overview

This project implements a modular pipeline for building, evaluating, and ensembling recommendation models using PyTorch and geometric approaches. Note that the core logic is now fully contained in the provided Jupyter notebook (final_combined.ipynb), as requireed by the guidelines for CIL-FS2025.

Colab Link: https://colab.research.google.com/drive/15noB-aEWY_43C2eUd5SBAtembEjoIw78?usp=sharing

Prerequisites

• Python 3.7 or higher
• A virtual environment tool (e.g., uv, venv, or conda)

Setup

1. Clone the repository:

   git clone <repository_url>
   cd <repository_directory>

2. Create and activate a virtual environment (using uv):

   uv create env
   uv activate env      # on Linux/MacOS and Windows

   (Alternatively, use venv or conda if preferred.)
3. Install dependencies:

   pip install pandas numpy matplotlib torch torch-geometric torch-sparse scikit-learn

Running the Notebook

1. Launch Jupyter Notebook:

   jupyter notebook

2. Open final_combined.ipynb and run all cells. The code should execute successfully once all dependencies are installed.

Implemented Methods

In this notebook, we apply the following methods:

1. Optimal Rank-(k) SVD
2. Embedding Dot-Product Model
3. Iterative SVD
4. SVD++
5. NeuMF
6. GraphNeuMF
7. DMF
8. Ensemble strategies (simple average, weighted average, top-(k) weighted, deep stacking)

Notes

• Ensure you have sufficient memory for data loading and model training.
• For GPU acceleration, verify that your PyTorch installation includes CUDA support.