Tencent's CONCORD-Rec Weights Each Modality by Reliability in Diffusion-Based Recommendation
_reachsumit · x · 2026-11-20
Tencent's CONCORD-Rec is a reliability-conditioned discrete diffusion recommender for semantic-ID (SID) based multimodal generative recommendation.
- Problem: uniform fusion and unconstrained token gates propagate misleading content when audiovisual, visual, or text streams are missing, degraded, or semantically conflicting.
- Method: counterfactual catalog-partition utility supervision — evaluating out-of-fold teacher coalitions over the exact feasible catalog support and aggregating item posteriors across the target code partition at each unresolved position; these loss differences supervise an availability-masked tokenwise gate that allocates bounded modality residuals around a behavioral denoiser, with an exact behavioral fallback slot.
- Includes theoretical fixed-support ranking stability bounds under gate perturbations, plus benchmarks under controlled missingness, degradation, and conflicting signals.
More from Multimodal
- Suspected GPT-6 Astra leak: one-prompt iterative 3D scene edits with editable Blender files — ZeroStateReflex · 2026-11-20
- Image Models Now Ship Fine-tuned LLM Prompt Upsamplers — Why Not Fuse Them? — ostrisai · 2026-11-20
- How to make viral 2000s-style AI video game trailers: full workflow, prompts, models — Div_pradeep · 2026-11-20
- AI-generated meme: Jensen Huang-like figure flanked by 'AI race accelerates' newspaper — AIandDesign · 2026-11-20
- 'Straw': an AI-crafted atmospheric short made with ComfyUI and Claude — TheHollywoodGeek · 2026-11-20
- Fable 5.2 turns reference photos into photorealistic unlit UE assets — theshawwn · 2026-11-20