PhD Dissertation Proposal by DONG Viet Hoang | Hybrid Learning Approaches for Recommendation: From Collaborative Signals to Language Understanding

Hybrid Learning Approaches for Recommendation: From Collaborative Signals to Language Understanding

ABOUT THE TALK

Recommender systems play a vital role in modern digital platforms by helping users navigate vast information spaces. Central to their effectiveness is the ability to learn rich latent representations of users and items. However, traditional collaborative filtering methods often struggle with data sparsity and inscrutable latent spaces. To address these limitations, this dissertation explores novel representation learning strategies that integrate user reviews and leverage advances in contrastive learning and pre-trained language models. 

The first part of this work introduces a Review-centric Contrastive Alignment Framework for Recommendation. Unlike previous review-aware models that treat reviews as auxiliary features, ReCAFR incorporates reviews directly into the representation learning process. It aligns user, item, and review representations within a unified latent space through two self-supervised contrastive strategies. In the second part, the dissertation bridges the gap between traditional collaborative filtering and modern language modeling. A novel framework is proposed that reformulates user-item interactions and metadata as natural language sequences, allowing pre-trained language models to generate user and item embeddings. Together, these contributions advance the field by offering new paradigms for integrating linguistic and behavioral data, paving the way toward more robust, data-efficient, and interpretable recommender systems.

SPEAKER BIOGRAPHY

DONG Viet Hoang is a PhD candidate in Computer Science at the School of Computing and Information Systems at SMU, under the supervision of Prof. Fang Yuan and co-supervisor Prof. Hady W. Lauw. His research focuses on enhancing the real-world recommender systems.