Reproduce “Training-free Adjustable Polynomial Graph Filtering for Ultra-fast Multimodal 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
Yu-Seung Roh, Joo-Young Kim, Jin-Duk Park, Won-Yong Shin
MM-GF turns multimodal recommendation into graph filtering by building separate item-similarity graphs from interactions, text, and images, then filtering and combining them into user scores without iterative training. On Baby, it reaches NDCG@20 of 0.0522 versus 0.0427 for MGCN, about 22.25% higher, while processing takes 8.1 seconds versus 13 minutes 33 seconds of MGCN training; this fixed-split result does not show live updating.
Key insights
- Explicitly adjusting spectral bounds on graph operators is crucial for maintaining stable and interpretable frequency responses in polynomial graph filters, especially with asymmetric normalization.
- Training-free polynomial graph filtering can achieve competitive or superior recommendation accuracy with significantly reduced computational costs compared to complex, iteratively trained GCN-based MRSs.
- The inherent smoothing property of graph filtering provides strong robustness against cold-start conditions and noisy multimodal features, offering a stable performance advantage over many parameterized models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.