Researcher in recommender systems
Bucher Sahyouni
Bucher Sahyouni is a PhD candidate in Artificial Intelligence at the University of Surrey who passed his viva with minor corrections in September 2026. His work studies recommender systems, neural ranking, multimodal and graph learning, and transformer-based sequential modelling under sparse implicit feedback.
Featured work
Research projects
DAP - Differential Adjusted Parity
A differentiable adjusted parity loss for learning fair representations without adversarial training.
MuSTRec
A multimodal and sequential transformer-based recommender that models item structure alongside short- and long-term user preferences.
MuSICRec
A contrastive graph recommender using sequence-item views and ID-guided multimodal fusion to address sparse interaction histories.
DSL - Dual-scale Softmax Loss
A softmax loss that adapts competition within and across training instances to improve Top-K recommendation and robustness.
Writing
Recent articles
What my thesis taught me about recommender systems
A draft reflection on recommender systems research, evaluation, and the practical lessons of thesis work.
Why sparse implicit feedback is difficult
A draft explanation of why implicit-feedback recommendation is challenging when most user-item pairs are unobserved.
Popularity bias in recommender systems: a simple explanation
A draft plain-language explanation of popularity bias and why exposure matters in recommendation.