Collaborative Filtering via Group-Structured Dictionary Learning

Structured sparse coding and the related structured dictionary learning problems are novel research areas in machine learning. In this paper we present a new application of structured dictionary learning for collaborative filtering based recommender systems. Our extensive numerical experiments demonstrate that the presented technique outperforms its state-of-the-art competitors and has several advantages over approaches that do not put structured constraints on the dictionary elements.

💡 Research Summary

The paper introduces a novel collaborative‑filtering framework that integrates structured sparse coding with dictionary learning, termed Group‑Structured Dictionary Learning (GSDL). Traditional recommender systems, whether based on matrix factorization (MF) or neural collaborative filtering (NCF), typically learn latent user and item factors without imposing any relational constraints on the dictionary atoms that compose the latent space. This lack of structure can lead to over‑fitting, reduced interpretability, and sub‑optimal performance, especially when data are extremely sparse.

GSDL addresses these issues by enforcing a predefined group structure on the dictionary elements. Each atom belongs to a group, and the learning objective includes a group‑wise regularizer that encourages atoms within the same group to share similar directions while allowing atoms from different groups to diverge. Mathematically, for an input vector (x) (e.g., a user’s rating profile) the model seeks a sparse coefficient vector (\alpha) and a dictionary (D) that minimize

\