Discrete Diffusion Models (DDMs) have recently been introduced to recommendation systems, modeling user history as a token generation process via iterative denoising. However, while effective at capturing user-level sequential patterns, these methods often fail to explicitly integrate item-based collaborative filtering information, a critical component for accurate recommendation. This deficiency manifests in two key aspects: (1) the item representation is often semantic-focused, lacking collaborative priors for diffusion training; and (2) the denoising process employs a uniform noise schedule, treating all tokens indiscriminately and ignoring item-level adaptive structural dependencies. To bridge this gap, we propose ANR-DiffRec, a unified framework designed to encode item-based collaborative structures into discrete diffusion for generative recommendation. First, we explicitly incorporate an item co-occurrence matrix to guide semantic ID generation, providing a structured collaborative prior for discrete diffusion training. Second, we introduce an item-based adaptive noise rescheduling mechanism that dynamically adjusts denoising weights according to both local contextual recoverability and behavior-aware item dependencies. Specifically, the proposed strategy jointly models intra-item structural context and inter-item collaborative signals, enabling structure-aware denoising during diffusion training. Extensive experiments on multiple benchmarks demonstrate that our method consistently outperforms state-of-the-art generative recommendation models. Code: https://github.com/CalmaQi/ANR-DiffRec.
Figures & tables
Figure 1. Illustration of the proposed item-based adaptive noise rescheduling strategy. (a) Existing diffusion-based recommendation models employ a uniform noise schedule, assigning identical denoising weights to masked tokens regardless of their contextual dependencies. (b) Our method dynamically adjusts the denoising weight for each token by jointly considering two complementary signals: (1) local recoverability estimation, and (2) behavior-aware dependency modeling. The final denoising weight is obtained by combining these two signals, enabling structure-aware diffusion training.
Figure 2. The architecture of ANR-DiffRec. (a) SID generation via fusion of semantic and collaborative signals. (b) Discrete diffusion training enhanced by item-based adaptive noise rescheduling, enabling structure-aware denoising to capture collaborative filtering signals. (c) Constrained discrete diffusion inference with validity filtering.
Dataset
#Users
#Items
#Interactions
Sparsity
Avg.Len
Scientific
50,985
25,848
412,947
99.969%
8.10
Instrument
57,439
24,587
511,836
99.964%
8.91
Game
94,762
25,612
814,586
99.966%
8.60
Steam
39,795
9,265
2,949,605
99.200%
74.12
MovieLens
6,040
3,883
1,001,456
95.730%
165.57
Table 1. Statistics of the used datasets. “Avg.Len” indicates the average number of interactions per user.
Category
Model
Scientific
Instrument
Game
R@1
R@5
R@10
N@5
N@10
R@1
R@5
R@10
N@5
N@10
R@1
R@5
R@10
N@5
N@10
Item ID based
GRU4Rec ( 13 )
0.0071
0.0184
0.0272
0.0128
0.0156
0.0094
0.0297
0.0453
0.0196
0.0246
0.0149
0.0461
0.0712
0.0307
0.0387
SASRec ( 20 )
0.0063
0.0240
0.0379
0.0152
0.0197
0.0089
0.0331
0.0525
0.0211
0.0273
0.0128
0.0516
0.0823
0.0323
0.0421
BERT4Rec ( 37 )
0.0045
0.0157
0.0264
0.0100
0.0134
0.0065
0.0255
0.0412
0.0160
0.0211
0.0082
0.0315
0.0530
0.0199
0.0267
FMLP-Rec ( 54 )
0.0046
0.0181
0.0300
0.0113
0.0151
0.0086
0.0299
0.0496
0.0193
0.0257
0.0099
0.0395
0.0649
0.0246
0.0328
LRURec ( 50 )
0.0049
0.0169
0.0267
0.0110
0.0141
0.0071
0.0272
0.0431
0.0172
0.0223
0.0134
0.0480
0.0753
0.0308
0.0396
Table 2. Performance comparison on three Amazon datasets. R@ K and N@ K denote Recall@ K and NDCG@ K , respectively. The best results are highlighted in bold , and the second-best results are underlined . The improvements of our model over the best baseline are statistically significant (paired t-test, p<0.05 ).
Dataset
Model
NDCG@5
NDCG@10
Recall@5
Recall@10
MovieLens
PreferGrow ( 18 )
0.0910
0.1177
0.1409
0.2237
RPG ( 17 )
0.1197
0.1440
0.1750
0.2503
DiffGRM ( 28 )
0.1038
0.1260
0.1541
0.2235
LLaDA-Rec ( 35 )
0.1359
0.1628
0.1980
0.2820
ANR-DiffRec
0.1468
0.1738
0.2141
0.2977
Steam
PreferGrow ( 18 )
0.0406
0.0508
0.0615
0.0935
Table 3. Performance comparison on MovieLens and Steam. The best results are highlighted in bold , and the second-best results are underlined . The improvements of our model over the best baseline are statistically significant (paired t-test, p<0.05 ).
Figure 3. Performance comparison across different user interaction sequence length groups on the Scientific and Instrument dataset.
Figure 4. Performance comparison across Head, Mid, and Tail item groups on the Scientific and Instrument dataset.
Dataset
Method
NDCG@1
NDCG@5
NDCG@10
Recall@5
Recall@10
Scientific
RQ-VAE
0.0120
0.0221
0.0269
0.0320
0.0487
w/o II
0.0109
0.0217
0.0268
0.0314
0.0479
w/o Schedule
0.0099
0.0205
0.0259
0.0312
0.0477
w/o Schedule and II
0.0097
0.0203
0.0253
0.0310
0.0474
Full
0.0122
0.0226
0.0279
0.0327
0.0494
Instrument
RQ-VAE
0.0128
0.0270
0.0341
0.0408
0.0626
Table 4. Ablation study results on three Amazon datasets. “II” denotes item co-occurrence information, and “Schedule” denotes the diffusion adaptive noise rescheduling strategy.
Figure 5. Ablation study on different noise rescheduling strategies.
Figure 6. Ablation study on different forms of collaborative signal construction.
Figure 7. Hyper-parameter study with different dimensionalities for collaborative embedding.
Figure 8. Hyper-parameter study of the sharpness in the geometric distribution.
Stage
Method
Time/Epoch (s)
Total (h)
Cost
Item Tokenization
RQ-VAE
0.648
1.799
1.000 ×
Ours
–
0.106
0.059 ×
Generative Training
LLaDA-Rec
87.402
3.641
1.000 ×
ANR-DiffRec
88.351
3.681
1.011 ×
Table 5. Training efficiency comparison in Scientific.
Method
Metric
Beam=1
Beam=10
Beam=20
Beam=30
Beam=40
Beam=50
LLaDA-Rec
Total Time (s)
0m 13s
1m 46s
3m 22s
5m 04s
6m 46s
8m 33s
Time / User (ms)
0.25
2.08
3.96
5.96
7.96
10.06
ANR-DiffRec
Total Time (s)
0m 13s
1m 46s
3m 24s
5m 06s
6m 50s
8m 39s
Time / User (ms)
0.25
2.08
4.00
6.00
8.04
10.18
Table 6. Inference efficiency comparison with different beam sizes in Scientific.
Table 7. Case study on three sampled items. We compare the number of ground-truth co-occurring items found within the Top-20 nearest neighbors in both the embedding space and the SID space. “Pre” and “Post” denote before and after integrating the item co-occurrence features.
Generative recommendation models user histories and item catalogs to generate personalized items, lists, or queries rather than only scoring candidates.
Recently, Generative Recommenders (GRs) have emerged as a transformative recommendation paradigm by replacing traditional item IDs with semantic indices (SIDs). Owing to the exceptional generative capabilities of diffusion models, a few pioneering works explore developing GRs with diffusion architectures as the backbone. However, a fatal limitation of existing diffusion-based GRs is that the diffusion process applies uniformly to all items within the historical interactions. In contrast, the user preference is shaped by multifaceted time-evolving factors and thus exhibits a non-stationary distribution in the temporal aspect. To bridge this gap, this study proposes a novel GR framework, named TDPM, by designing the time-aware diffusion on SID tokens. Specifically, TDPM explicitly integrates the impact of time-evolving user preferences into the diffusion process. In detail, the user preference is disentangled into (i) the period preference, which remains consistent over a long time-span, and (ii) the point preference, which is triggered by recent focal events. Extensive experiments on three public real-world datasets demonstrate the significant superiority of TDPM over the state-of-the-art baselines. TDPM achieves average improvements of up to 29.21% and 25.45% in terms of HR@20 and NDCG@20, respectively. The ablation study further underscores the necessity of time-aware token diffusion in diffusion-based GRs.
Bangguo Zhu, Peng Huo, Yuanbo Zhao +3
Central South University Changsha, China · National Super Computing Center Tianjin, China · Renmin University of China Beijing, China +1
Cold-start item recommendation remains a persistent challenge in real-world systems due to the absence of interaction histories. While prior models attempt to bridge this gap using item content features, they universally suffer from the seesaw dilemma: enhancing performance for cold items inevitably degrades performance for warm items, and vice versa. We identify that this dilemma stems from a fundamental distributional disparity: warm item embeddings occupy a complex behavioral manifold" shaped by rich interaction signals, whereas cold item embeddings are constrained to a semantic manifold" derived solely from auxiliary content. Existing methods often force a rigid mapping between these inconsistent spaces, causing the model to sacrifice the precision of warm representations to accommodate cold ones. To address this, we propose DiffCold, a diffusion-based generative model that unifies warm and cold representations. Unlike GANs or VAEs, DiffCold leverages conditional diffusion to reconstruct warm item embeddings from content, preserving the underlying manifold structure without degradation. We further tailor this paradigm with two specific designs: a Retrieval-enhanced Aggregator that initializes generation using semantically similar warm items to bypass inefficient noise, and a Simulation-based Representation Alignment module that enforces distribution consistency between generated and real embeddings via contrastive learning. Experiments on three benchmarks confirm that DiffCold resolves the seesaw dilemma, consistently outperforming state-of-the-art methods across all metrics.
Kangning Zhang, Yingjie Qin, Weinan Zhang +2
Shanghai Jiao Tong University, Shanghai, China · Xiaohongshu Inc., Shanghai, China
Collaborative filtering (CF) recommendation has been significantly advanced by integrating Graph Neural Networks (GNNs) and Graph Contrastive Learning (GCL). However, (i) random edge perturbations often distort critical structural signals and degrade semantic consistency across augmented views, and (ii) data sparsity hampers the propagation of collaborative signals, limiting generalization. To tackle these challenges, we propose RaDAR (Relation-aware Diffusion-Asymmetric Graph Contrastive Learning Framework for Recommendation Systems), a novel framework that combines two complementary view generation mechanisms: a graph generative model to capture global structure and a relation-aware denoising model to refine noisy edges. RaDAR introduces three key innovations: (1) asymmetric contrastive learning with global negative sampling to maintain semantic alignment while suppressing noise; (2) diffusion-guided augmentation, which employs progressive noise injection and denoising for enhanced robustness; and (3) relation-aware edge refinement, dynamically adjusting edge weights based on latent node semantics. Extensive experiments on three public benchmarks demonstrate that RaDAR consistently outperforms state-of-the-art methods, particularly under noisy and sparse conditions.
Yixuan Huang, Jiawei Chen, Shengfan Zhang +1
National University of Defense Technology, Changsha, China · Ant Group, Chongqing, China · Tsinghua University, Beijing, China