Reproduce “Do Recommender Systems Really Leverage Multimodal Content? A Comprehensive Analysis on Multimodal Representations for Recommendation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Claudio Pomo, Matteo Attimonelli, Danilo Danese, Fedelucio Narducci, Tommaso Di Noia
The authors test whether meaningful item features help multimodal recommendation beyond input dimensionality by comparing vision-language-model attributes and visual embeddings with matched noise. For VBPR on Pets, ResNet50 features yield 9.53% Recall@20, versus 8.58% and 8.43% for Gaussian and multivariate noise. This supports informative inputs in this setting, not semantics as the sole cause; BM3’s gap is smaller.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.