Research map / Vision-language and multimodal / Cross-modal learning

Curriculum Conditioned Diffusion for Multimodal Recommendation

AAAI 2025 · Vision-language and multimodal › Cross-modal learning · approach: Multimodal sentiment analysis · 8 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes CCDRec, a diffusion-based multimodal recommender that aligns multimodal and collaborative item representations and uses the reverse diffusion process to build a curriculum of progressively harder negative samples.

Problem

Multimodal recommendation overlooks negative instances, and existing negative sampling strategies are designed for collaborative filtering rather than multimodal representations.

Approach

Uses a conditional DDPM-style diffusion module (DMA) to align multimodal features with collaborative signals, then samples negatives from intermediate reverse-diffusion steps (NDI) on a curriculum schedule (CNS) that moves from easy to hard negatives.

Key results

• CCDRec(FREEDOM) reaches Recall@10 0.0679 on Baby vs 0.0626 for FREEDOM (about 8.5% relative gain).
• On Clothing, CCDRec(MG) improves NDCG@10 from 0.0336 to 0.0361 and CCDRec(LATTICE) improves Recall@10 from 0.0499 to 0.0613 relative to its backbones.
• CCDRec outperforms DNS, MixGCF and RealHNS when each is added to LATTICE, FREEDOM and MG on Baby and Clothing (e.g., Clothing Recall@10 0.0677 vs 0.0586 best HNS baseline with FREEDOM).

Measured effects reported in the paper

Limitations

• Gains over backbones are modest on some metrics and datasets.
• Evaluated only on three Amazon datasets with ID and pre-extracted visual/text features.

Experimental setup

Datasets: Amazon Baby Amazon Sports Amazon Clothing

Benchmarks: Amazon Baby Amazon Sports Amazon Clothing

Metrics: Recall@5 Recall@10 NDCG@5 NDCG@10

Baselines: BPR-MF LightGCN MMGCN SLMRec BM3 LATTICE FREEDOM MCDRec MG DNS MixGCF RealHNS

Builds on: FREEDOM DDPM Curriculum learning BPR DiffRec

Closest papers by meaning

• Seeing Beyond Noise: Joint Graph Structure Evaluation and Denoising for Multimodal Recommendation (AAAI 2025 · similarity 0.90)
• MENTOR: Multi-level Self-supervised Learning for Multimodal Recommendation (AAAI 2025 · similarity 0.90)
• LGMRec: Local and Global Graph Learning for Multimodal Recommendation (AAAI 2024 · similarity 0.89)
• Beyond Graph Convolution: Multimodal Recommendation with Topology-aware MLPs (AAAI 2025 · similarity 0.88)
• Plug-In Diffusion Model for Sequential Recommendation (AAAI 2024 · similarity 0.88)
• Structured Spectral Reasoning for Frequency-Adaptive Multimodal Recommendation (NeurIPS 2025 · similarity 0.87)