A Comparative Study on the Feasibility of Graph-based Recommendation Systems

Authors: Sharjeel Mustafa & George Kaceli

Introduction

In the digital age, attention is a key driver of economic success. Whether it’s a poster, a product, or a book, capturing a potential customer’s attention can be the deciding factor in whether they choose to engage or walk away. As a result, recommender systems play a critical role in personalizing content per user and converting them into a sale. In this work, we investigate a broad spectrum of recommender system designs, focusing on the feasibility of leveraging graph-based approaches to more effectively utilize relational information. Recommender systems are typically categorized into three main types: content-based, collaborative filtering, and hybrid models. We evaluate a diverse range of graph-based models alongside established methods from each category. Notably, the graph-based NGCF model slightly outperforms its non-graph-based counterparts, even when applied to data that is not inherently graph-structured, thereby demonstrating the broader applicability and potential of graph-based methods in general recommendation scenarios.

Citation

If you find this work useful, please consider citing:

@misc{sharjeelm,
  title={A Comparative Study on the Feasibility of Graph-based Recommendation Systems},
  author={Sharjeel Mustafa and George Kaceli},
  year={2025},
  url={https://github.com/Sharjeeliv/C874-final/}
}