RoFiGR injects collaborative signals into coarse SID routing for multimodal generative recommenders
_reachsumit · x · 2026-10-30
RoFiGR addresses early routing errors in multimodal generative recommenders, where a wrong coarse SID code can exclude relevant items before fine-grained prediction. Its two-tree recommender uses collaborative information in both SID construction and generation: aligning cumulative L1-L2 code vectors with frozen collaborative item embeddings, encoding item distance/source level in serialized history, and supervising legal branches with a sequential teacher plus bounded adaptive residual weighting. It beats external baselines on all five ranking metrics across Instruments, Arts, and Games, with five-seed paired tests (Holm-corrected).
More from Research
- NBER opens applications for Stanford Digital Economics and AI tutorial, due Nov 3 — avicgoldfarb · 2026-10-30
- FlexUMM shows unified multimodal models aren't so unified in embedding space — jindong_wang92 · 2026-10-30
- Data leakage hit 648 papers across 30 fields; autoarxiv now auto-reproduces arXiv papers — pandeyparul · 2026-10-30
- CMU paper: one algorithm for pretraining and post-training — unified training beats staged pipelines — heghbalz · 2026-10-30
- IDM Distillation Hits Best 2-NFE Long-Text Rendering: 94.25% CN / 90.20% EN — _akhaliq · 2026-10-30
- Neural CPU runs Doom E1M1: 28.7M tokens, 2h21m for a verified real frame — SalamanderFickle4315 · 2026-10-30