Description:

Recommender systems are algorithms designed to suggest items—such as videos, books, or tweets—to users based on their preferences. Among various approaches, matrix factorization is one of the most widely used techniques. In this method, the user-item interaction matrix (where each element represents the latent rating or preference of a user for an item) is assumed to exhibit a low-rank structure.

In this project, we aim to propose further refining this assumption by exploring hierarchical structures within the user-item matrix. We aim to investigate whether incorporating hierarchical relationships can enhance the performance of recommender systems.

Additionally, we seek to develop a more robust recommender system that is resilient to the influence of malicious users, thereby improving the reliability and trustworthiness of recommendations.

Skills:

• Proficiency in Python
• Linear Algebra
• Basic knowledge of recommender systems is a plus
• Basic Knowledge of clustering is a plus

Relevant Resources:

• Trust-Aware Collaborative Filtering for Recommender Systems 2004
• Probabilistic Matrix Factorization 2007

How to Apply:

Send your CV and transcript to Oscar and Daichi at oscar.villemaud@epfl.ch and daichi.kuroda@epfl.ch