MDVT: Enhancing Multimodal Recommendation with Model-Agnostic Multimodal-Driven Virtual Triplets

Edith C. H. Ngai; Hewei Wang; Jinfeng Xu; Jinze Li; Mengran Li; Puzhen Wu; Shuo Yang; Yijie Li; Zheyu Chen

REVIEW 4 major objections 6 minor 3 cited by

MDVT: Enhancing Multimodal Recommendation with Model-Agnostic Multimodal-Driven Virtual Triplets

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper shows that cosine similarity between user and item multimodal embeddings can generate high-quality virtual training triplets, and that adding them to the training loss improves recommendation accuracy, most for users with hardly…

desk verdict A solid empirical plug-in idea for multimodal recommenders, but a cross-table metric inconsistency and unreleased code need fixing before I'd trust the universal gains. read the letter →

arxiv 2505.16665 v1 pith:6XVUZVY3 submitted 2025-05-22 cs.IR

Jinfeng Xu , Zheyu Chen , Jinze Li , Shuo Yang , Hewei Wang , Yijie Li , Mengran Li , Puzhen Wu

show 1 more author

Edith C. H. Ngai

This is my paper · ORCID

classification cs.IR

keywords RecommenderSystemMultimediaModel-AgnosticVirtualTripletsDataSparsityMultimodalRecommendationBPRLossWarm-upStrategies

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that the cosine similarity between a user's fused multimodal embedding and items' fused multimodal embeddings can be trusted, after a warm-up phase, to identify items the user would prefer and items they would not. On that basis it constructs virtual training triplets—virtual positives from the most similar items and virtual negatives from the least similar—and adds a Bayesian Personalized Ranking (BPR) loss over these triplets to the model's original loss. The paper reports that this plug-and-play addition improves Recall@K and NDCG@K across six existing multimodal recommendation models and four real-world datasets, with the largest relative gains for users who have interacted with very few items. If correct, the finding matters because it shows multimodal signals can supply supervision where interaction data are missing, without changing the underlying model architecture.

What carries the argument

The carrying mechanism is the multimodal-driven virtual triplet constructor: a running cosine-similarity matrix $S_{u,i} = \bar{\mathbf{e}}_u^\top \bar{\mathbf{e}}_i / (\|\bar{\mathbf{e}}_u\| \|\bar{\mathbf{e}}_i\|)$ between each user's fused embedding $\bar{\mathbf{e}}_u$ and each item's fused embedding $\bar{\mathbf{e}}_i$ (Eq. 5). When a warm-up threshold is reached, the top-$n$ and bottom-$n$ entries for each user become virtual positives and negatives (Eq. 6), and their averaged embeddings define the virtual positives and negatives used in the loss (Eq. 8). These virtual triplets are optimized with a weighted joint loss $(1-\lambda)\mathcal{L}_{\mathrm{bpr}} + \lambda \mathcal{L}_{\mathrm{vbpr}}$ (Eq. 9), where $\lambda$ balances the magnitude of the two losses to avoid gradient skew. Three warm-up threshold strategies—static grid search, dynamic loss-change detection, and a hybrid that narrows the grid around the dynamic estimate—decide when the virtual triplets are trustworthy enough to join the optimization.

What would settle it

Train the same models with MDVT but replace the similarity scores that build virtual triplets with random scores drawn from the same distribution. If the random-triplet version matches the similarity-triplet version in Recall@K and NDCG@K, then the alleged mechanism—multimodal similarity as supervision—is not driving the improvement.

Watch

Extended reading notes

Core claim

The central claim is that multimodal information can do more than enrich item representations: it can generate supervision for users. The paper argues that once a multimodal recommendation model has been warmed up to learn meaningful user and item embeddings, the cosine similarity between a user's fused embedding and every item's fused embedding ranks items by predicted affinity. Selecting the top-n and bottom-n items in this ranking as virtual positives and negatives creates a virtual triplet dataset, and training the model jointly on the real BPR loss and a BPR loss over these virtual triplets yields consistent gains in Recall@K and NDCG@K. The authors report that the gains hold across six published multimodal recommendation models on four datasets, are largest for sparse users, and that the virtual-triplet supervision also accelerates convergence. A notable supporting result is that using only visual or only textual modality to build the triplets underperforms the original model, while using ID alone already helps and fusing all modalities works best.

Load-bearing premise

After enough warm-up training, the cosine similarity between a user's fused multimodal embedding and an item's fused multimodal embedding ranks items the way the user would, so the most similar items can be treated as preferred and the least similar as non-preferred.

Editorial extensions

If this is right

Any multimodal recommendation model that produces a fused user embedding and a fused item embedding can be augmented with the virtual-triplet loss without architectural changes.
The method should give the largest gains on cold-start and extremely sparse users, since the virtual triplets compensate for missing interaction signals.
The hybrid warm-up strategy offers a practical way to avoid exhaustive hyper-parameter search while approaching the static strategy's performance.
The virtual-triplet loss can be combined with adversarial training and data augmentation, as the reported compatibility experiments show further gains.
Because the virtual triplets are rebuilt from evolving representations, the method adds no new data requirement beyond what the multimodal model already consumes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the similarity ranking is truly preference-aligned, the same virtual-triplet construction could be plugged into other pairwise or listwise ranking losses, not just BPR.
The approach implicitly assumes the fused embedding space is locally smooth; a direct test would be to measure how often held-out true positives appear among the virtual positives during training.
The dynamic warm-up criterion based on loss change could be replaced by a criterion based on embedding stability, which might be more reliable across optimizers and learning-rate schedules.
One could re-weight the virtual triplets by the confidence of the similarity ranking (for example, the margin between top-n and bottom-n similarities), potentially reducing noise from ambiguous users.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The genuinely new piece is building virtual triplets from the cosine similarity of dynamically learned user-item fused representations, after a warm-up, rather than from raw item features as CC-GCN and LATTICE do. The three threshold strategies (static, dynamic, hybrid) are a practical contribution, and the hybrid one sensibly cuts search cost. What the paper does well is empirical breadth: six baselines, four datasets, sparsity breakdowns, convergence curves, and hyperparameter analysis. The ablations are informative, and Appendix A.4 is honest about the failure mode without a warm-up, including the popularity bias risk for sparse users. That is the kind of self-criticism that makes me trust the core result more.

The soft spots are real but addressable. First, Table 4's MMGCN/Clothing NDCG@5 origin cell reads 0.0110, while Table 2 lists 0.0088; the +M value 0.0099 is below 0.0110 but above 0.0088. Read literally, that cell contradicts Observation 1's universal improvement claim. The most plausible reading is a swapped column: 0.0110 is Table 2's NDCG@10 value, so the Table 4 origin cell was probably meant to be 0.0088. Still, this must be fixed; a single mislabeled cell in the central tables casts doubt on the rest. Second, the code is only footnote-referenced, so the numbers cannot be independently checked. Third, the static threshold search uses validation performance to pick warm-up epochs, and the paper does not state whether the same validation split was then reused for early stopping or checkpoint selection; threshold leakage is not obvious but should be clarified.

The circularity concern is minor. Yes, the labels come from the model's own representations, but the paper does not pretend otherwise, and the warm-up plus the frequency-prior ablations give the claim some teeth. I would not call the method theoretically unsound; I would call it a self-training procedure with an empirically justified schedule.

Bottom line: this is a useful paper for anyone working on multimodal recommendation or sparsity. It deserves peer review, but the authors should release runnable code, fix the Table 4 inconsistency, and state the validation protocol for threshold selection. If those are addressed, I would accept.

Referee Report

4 major / 6 minor

Summary. MDVT is a model-agnostic auxiliary-training method for multimodal recommendation. After a warm-up period, it constructs virtual triplets by taking, for each user, the n most similar and n least similar items according to cosine similarity over the model's fused user/item representations (Eqs. 5-6). A BPR loss over these virtual triplets is added to the original BPR loss with a balancing weight (Eqs. 7-9). Three warm-up threshold strategies are proposed: static, dynamic, and hybrid. The paper reports experiments on Baby, Sports, Clothing, and TikTok across six multimodal baselines, plus ablations, sparsity analysis, convergence curves, and compatibility with adversarial training and GPT-4o augmentation. The central claim is that MDVT improves all tested baselines across all datasets and metrics, with the largest relative gains for sparse users.

Significance. If the reported results are reliable, MDVT is a simple and potentially useful plug-and-play component for multimodal recommenders, and the extensive evaluation across six baselines and four datasets is a strength. The paper also includes useful internal analyses that go beyond a single headline number: the ablation of triplet sources (Table 3), the sparsity study (Figure 4), the convergence study (Figure 5), and the honest Appendix A.4 discussion of what happens when the warm-up assumption fails. The main significance depends entirely on the correctness of the reported tables, and one table contains an internal inconsistency that must be resolved before the empirical claim can be accepted.

major comments (4)

[Table 4 vs Table 2] Table 4 reports for MMGCN on Clothing an origin NDCG@5 of 0.0110, while Table 2 reports the same condition (MMGCN, Clothing, N@5) as 0.0088. The +M (i.e., +MDVT) value in Table 4 is 0.0099, which is below the Table 4 origin value. Read literally, this cell shows MDVT reducing NDCG@5 on Clothing, contradicting Observation 1 in Section 4.2 that all warm-up threshold strategies improve over all baselines across all evaluation metrics. The Table 4 origin value also coincides with the Table 2 N@10 value for MMGCN/Clothing (0.0110), which suggests a column mislabeling; however, as printed the two tables cannot both be correct. This discrepancy must be corrected and the affected performance claims re-verified, because the paper's central empirical claim rests on these tables.
[Section 3.1, Eqs. (5)-(6)] The construction of virtual triplets does not explicitly exclude items already recorded in the interaction matrix R. If an item with R_{u,i}=1 is among the top-n most similar items for user u, then the virtual positive group in Eq. (8) contains an observed positive, and the virtual BPR loss in Eq. (7) partly re-weights existing supervision rather than providing genuinely new supervision. Since the paper's premise is that virtual triplets supply signal beyond observed interactions, the authors should either exclude observed positives (and, ideally, observed negatives) before the top/bottom-n selection, or report the degree of overlap. A variant that removes observed interactions before building D^V would directly test whether the reported gains come from new pairwise supervision.
[Section 3.3, Eqs. (7)-(9)] The virtual labels are endogenous: the same fused representations are used both to select virtual positives/negatives and to compute the BPR loss that updates those representations. This is a self-training loop, and the paper's main safeguard is the warm-up threshold. Appendix A.4 itself shows that the loop fails without a warm-up and can amplify popularity bias for users with 1-3 interactions. To make the central claim robust, the authors should give a concrete evaluation of the loop's reliability, for example: (i) measure the precision of top-n virtual positives against held-out interactions during the warm-up period; (ii) report the overlap between virtual positives and popular items; or (iii) compare against a control loss using random triplets with the same number of extra pairs. The random-triplet control is especially important because the observed gains could otherwise be attributable to the extra regularization from additional training pairs rather than to multimodal-driven selection.
[Section 4.1 and Table 2 significance asterisks] Many Table 2 improvements are marked with p<0.01, but the paper does not state the number of random seeds, the statistical test used, or where variances are reported. Because the universal-improvement claim is the paper's main result, the testing procedure should be specified (e.g., paired test over seeds, number of seeds, and standard deviations in an appendix or supplementary table). Without this information, the significance asterisks cannot be independently assessed.

minor comments (6)

[Section 3, first paragraph] The text refers to 'the overall framework of our proposed MDVR', which should read MDVT.
[Section 4.1.2] The text says the experiments examine 'five multimodal recommendation models', but Table 2 reports results for six models, including MMSSL.
[Section 4.8] The sentence 'for MMGCN and SLMRec, the optimal hyper-parameters are higher, with λ = 0.2 and n = 4' is internally inconsistent for λ, since 0.2 is the same value given for the other models.
[Appendix A.1] The text says the code link can be found in a footnote, but no footnote or URL is visible in the arXiv version; the link should be included explicitly.
[Appendix A.4] The abbreviations 'MDVT (w p w/o n)', 'MDVT (w p)', and 'MDVT+ (w p)' are used in Tables 6 and 7 and in the discussion but are not defined with enough clarity in the main text; please spell out the variants explicitly at first use.
[Algorithm 1] In Algorithm 1, the update 'Update L_prev by current loss L' appears inside the loop but L_prev is not initialized before the first epoch; the pseudocode should clarify the initial condition.

Circularity Check

2 steps flagged · score 5.0 of 10

Virtual-triplet labels are constructed from the model's own similarity ranking and then reinforced by the same BPR objective (Eqs. 5-7), and Algorithm 1 selects the warm-up threshold on the reported test metric; both steps make part of the claimed gain endogenous.

self definitional [Section 3.1 Eq. (5)-(6) and Section 3.3 Eq. (7)-(9)]
"we select the most similar n and the least similar n items to construct the virtual triplet: D^V_{u,i+}=Max_n(S_{u,i*}|i*∈I), D^V_{u,i-}=Max_n(−S_{u,i*}|i*∈I) ... L_vbpr = Σ_{（u,D^V_{u,i+},D^V_{u,i-}）∈D^V} −log(σ(\bar e_u^⊤ \hat e_{u,i+} − \bar e_u^⊤ \hat e_{u,i-}))"

The virtual 'positive' and 'negative' items are defined as the items with the highest and lowest cosine similarity S_{u,i} under the current fused representations (Eq. 5-6). The virtual BPR loss (Eq. 7) then trains the model to increase exactly that same similarity score for the chosen positives and decrease it for the chosen negatives. Thus the auxiliary supervision signal is the model's own ranking reinjected as ground truth; any improvement from L_vbpr is partly self-reinforcement rather than new information from multimodal content. The paper's own Appendix A.4 concedes that without a warm-up the model selects 'incorrect high-similarity items as virtual triplets,' confirming that the labels are endogenous and their quality depends on the very model being trained.
fitted input called prediction [Appendix A.1 Algorithm 1, static/hybrid branch, lines 41-44]
"Test model performance P; if P > P_bar then Update P_bar = P, T_cur = T_S"

The static and hybrid warm-up threshold strategies select the number of warm-up epochs by evaluating on the test set ('Test model performance P') and keeping the configuration with the highest test metric. The same test metrics are then reported in Table 2 as evidence that MDVT 'significantly improves over all baselines across all evaluation metrics.' This means the reported improvement for the static/hybrid strategies is the maximum of the evaluation metric over the searched warm-up epochs on the test set, so the 'prediction' of improvement is partly a test-set selection artifact rather than an independent evaluation. The paper does not state that a separate validation set is used for this selection, despite Section 4.1.1 mentioning an 8:1:1 train/validation/test split.

full rationale

The core MDVT mechanism is a self-training loop: virtual positives and negatives are defined by the model's current user-item similarity (Eq. 5-6), and the auxiliary loss (Eq. 7) is a BPR objective over those same similarities. This is not an externally grounded supervision signal, so the observed gains are at least partly self-reinforcement. In addition, Algorithm 1 selects the warm-up threshold by test performance and then reports that same test performance, which makes the static/hybrid results statistically optimistic. These two issues make the central empirical claim partially circular. However, the paper does evaluate final models on held-out test data against standard baselines, and there is no load-bearing self-citation or imported uniqueness theorem; the related-work self-citations (e.g., COHESION, MENTOR, the authors' own survey) are not used to justify the MDVT mechanism. A separate, non-circularity concern is the internal numerical inconsistency between Table 2 and Table 4 for Clothing/MMGCN NDCG@5 (Table 2 baseline 0.0088, +MDVT(H) 0.0099; Table 4 'origin' 0.0110, +M 0.0099, which would imply MDVT hurts), and the code is only a non-inspectable footnote reference. That is a data-integrity and reproducibility issue, not a derivation circularity, but it compounds the reliability risk of the headline claim.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

No new physical or conceptual entity is postulated. Virtual triplets are a training construct built from existing embeddings, not an independent falsifiable entity. The central claim rests on five fitted hyperparameters and four domain assumptions about representation similarity and encoder quality.

free parameters (5)

lambda (virtual loss weight) = 0.2 for most models and datasets
Eq. 9 balances L_bpr and L_vbpr; grid searched over {0.1,0.2,0.3,0.4,0.5} in Section 4.1.3 and analyzed in Section 4.8.
n (number of virtual positives/negatives per user) = 2 for FREEDOM/DRAGON/LGMRec, 4 for MMGCN/SLMRec
Top-n filter in Eq. 6; grid searched over {1,2,4,8}.
Static warm-up threshold T_S = Per model and dataset, from {0,5,10,20,40,80}
Determines when virtual triplets join training; exhaustively searched under the static strategy.
Dynamic loss-change threshold g = 0.1 or 0.2
Stops warm-up when relative BPR loss decrease falls below g; grid searched over {0.1,0.2,0.3,0.4}.
Hybrid search scope s = 2
Static refinement window [T_cur-s, T_cur+s] around dynamic estimate; grid searched over {1,2,3,4,5}.

assumptions (4)

domain assumption Cosine similarity between a user's fused embedding and item embeddings becomes a trustworthy ranking signal after warm-up, so top-n similar items are valid virtual positives and bottom-n are valid virtual negatives.
Core premise of Section 3.1 (Eq. 5-6); only empirically validated, and Appendix A.4 shows it fails without warm-up.
domain assumption User modality embeddings initialized randomly and refined by graph propagation are comparable to pretrained item modality embeddings in the fused space.
Section 2 and Section 3.1: users have no raw multimodal features, yet their learned embeddings are used in cosine similarity against item embeddings.
standard math The BPR pairwise ranking objective and cosine similarity are appropriate for the fused multimodal representation space.
Used in Eq. 3, Eq. 5, and Eq. 7; standard in collaborative filtering but adopted without justification.
domain assumption Pretrained visual and textual encoders produce item features that are informative for the recommendation task.
Item embeddings are extracted by pretrained encoders (Section 2); the paper does not validate encoder quality or modality alignment.

how reviews work

0 comments

Cite this review

Pith. "Pith review of MDVT: Enhancing Multimodal Recommendation with Model-Agnostic Multimodal-Driven Virtual Triplets." pith.science (2026). https://pith.science/paper/6XVUZVY3

@misc{pith2026250516665,
  author       = {Pith},
  title        = {Pith review of: MDVT: Enhancing Multimodal Recommendation with Model-Agnostic Multimodal-Driven Virtual Triplets},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6XVUZVY3}},
  note         = {Machine review of arXiv:2505.16665}
}

read the original abstract

The data sparsity problem significantly hinders the performance of recommender systems, as traditional models rely on limited historical interactions to learn user preferences and item properties. While incorporating multimodal information can explicitly represent these preferences and properties, existing works often use it only as side information, failing to fully leverage its potential. In this paper, we propose MDVT, a model-agnostic approach that constructs multimodal-driven virtual triplets to provide valuable supervision signals, effectively mitigating the data sparsity problem in multimodal recommendation systems. To ensure high-quality virtual triplets, we introduce three tailored warm-up threshold strategies: static, dynamic, and hybrid. The static warm-up threshold strategy exhaustively searches for the optimal number of warm-up epochs but is time-consuming and computationally intensive. The dynamic warm-up threshold strategy adjusts the warm-up period based on loss trends, improving efficiency but potentially missing optimal performance. The hybrid strategy combines both, using the dynamic strategy to find the approximate optimal number of warm-up epochs and then refining it with the static strategy in a narrow hyper-parameter space. Once the warm-up threshold is satisfied, the virtual triplets are used for joint model optimization by our enhanced pair-wise loss function without causing significant gradient skew. Extensive experiments on multiple real-world datasets demonstrate that integrating MDVT into advanced multimodal recommendation models effectively alleviates the data sparsity problem and improves recommendation performance, particularly in sparse data scenarios.

Figures

Figures reproduced from arXiv: 2505.16665 by the authors.

**Figure 2.** Ablation study on key components of MDVT in terms of NDCG@10. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** The learning curve when adopting MDVT and its [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Sparsity study on three advanced multimodal rec [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Convergence study on the Baby dataset [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: Mechanics of all three warm-up threshold strategies for two advanced models on the Baby dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

**Figure 7.** Figure 7: Performance 𝑤 .𝑟 .𝑡 . hyper-parameter 𝜆. 1 2 4 8 Hyper-parameter n 0.04 0.06 0.08 Recall@10 Baby 1 2 4 8 Hyper-parameter n Sports 1 2 4 8 Hyper-parameter n Clothing MMGCN SLMRec FREEDOM DRAGON LGMRec [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

The Best is Yet to Come: Graph Convolution in the Testing Phase for Multimodal Recommendation
cs.IR 2025-07 conditional novelty 6.0 of 10

A multimodal recommender that trains without graph convolution and applies it only at test time outperforms graph-trained baselines while training much faster.
NLGCL: Naturally Existing Neighbor Layers Graph Contrastive Learning for Recommendation
cs.IR 2025-07 conditional novelty 6.0 of 10

NLGCL treats each user/item embedding and its graph neighbors at the next GNN layer as positive pairs, eliminating augmentation-based contrastive views in GNN recommenders while improving accuracy and training speed.
Leveraging Optimal Transport for Distributed Two-Sample Testing: An Integrated Transportation Distance-based Framework
stat.ME 2025-06 reject novelty 4.0 of 10

A permutation test that aggregates per-client Wasserstein distances into an integrated transportation distance detects distributional differences in distributed data, with claimed Type I error control and high power.

Reference graph

Works this paper leans on

54 extracted references · 16 canonical work pages · cited by 3 Pith papers

[1]

Léon Bottou, Frank E Curtis, and Jorge Nocedal. 2018. Optimization methods for large-scale machine learning.SIAM review60, 2 (2018), 223–311

2018
[2]

Jiaben Chen, Xin Yan, Yihang Chen, Siyuan Cen, Qinwei Ma, Haoyu Zhen, Kaizhi Qian, Lie Lu, and Chuang Gan. 2024. RapVerse: Coherent Vocals and Whole-Body Motions Generations from Text.arXiv preprint arXiv:2405.20336(2024)

work page arXiv 2024
[3]

Jingyuan Chen, Hanwang Zhang, Xiangnan He, Liqiang Nie, Wei Liu, and Tat- Seng Chua. 2017. Attentive collaborative filtering: Multimedia recommendation with item-and component-level attention. InProceedings of the 40th International ACM SIGIR conference on Research and Development in Information Retrieval. 335–344

work page 2017
[4]

Xu Chen, Hanxiong Chen, Hongteng Xu, Yongfeng Zhang, Yixin Cao, Zheng Qin, and Hongyuan Zha. 2019. Personalized fashion recommendation with visual explanations based on multimodal attention network: Towards visually explainable recommendation. InProceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval...

2019
[5]

Zheyu Chen, Jinfeng Xu, and Haibo Hu. 2025. Don’t Lose Yourself: Boosting Multimodal Recommendation via Reducing Node-neighbor Discrepancy in Graph Convolutional Network. InICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 1–5

work page 2025
[6]

Zheyu Chen, Jinfeng Xu, Yutong Wei, and Ziyue Peng. 2025. Squeeze and Excita- tion: A Weighted Graph Contrastive Learning for Collaborative Filtering.arXiv preprint arXiv:2504.04443(2025)

arXiv 2025
[7]

Zhiqiang Guo, Jianjun Li, Guohui Li, Chaoyang Wang, Si Shi, and Bin Ruan. 2024. LGMRec: Local and Global Graph Learning for Multimodal Recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 8454–8462

work page 2024
[8]

Ruining He and Julian McAuley. 2016. VBPR: visual bayesian personalized ranking from implicit feedback. InProceedings of the AAAI conference on artificial intelligence, Vol. 30

2016

Show all 54 references

[9]

Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 639–648

2020
[10]

Feiran Huang, Zhenghang Yang, Junyi Jiang, Yuanchen Bei, Yijie Zhang, and Hao Chen. 2024. Large Language Model Interaction Simulator for Cold-Start Item Recommendation.arXiv preprint arXiv:2402.09176(2024)

2024 arXiv
[11]

Yangqin Jiang, Lianghao Xia, Wei Wei, Da Luo, Kangyi Lin, and Chao Huang
[12]

Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic opti- mization.arXiv preprint arXiv:1412.6980(2014)

2014 arXiv
[13]

Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer, and Tom Goldstein. 2018. Visualizing the loss landscape of neural nets.Advances in neural information processing systems31 (2018)

2018
[14]

Ruirui Li, Xian Wu, and Wei Wang. 2020. Adversarial learning to compare: Self-attentive prospective customer recommendation in location based social networks. InProceedings of the 13th International Conference on Web Search and Data Mining. 349–357

2020
[15]

Zhiqiu Lin, Siyuan Cen, Daniel Jiang, Jay Karhade, Hewei Wang, Chancharik Mitra, Tiffany Ling, Yuhan Huang, Sifan Liu, Mingyu Chen, et al. 2025. Towards Understanding Camera Motions in Any Video.arXiv preprint arXiv:2504.15376 (2025)

2025 arXiv
[16]

Shang Liu, Zhenzhong Chen, Hongyi Liu, and Xinghai Hu. 2019. User-video co- attention network for personalized micro-video recommendation. InThe world wide web conference. 3020–3026

2019
[17]

Sichun Luo, Yuxuan Yao, Bowei He, Yinya Huang, Aojun Zhou, Xinyi Zhang, Yuanzhang Xiao, Mingjie Zhan, and Linqi Song. 2024. Integrating large language models into recommendation via mutual augmentation and adaptive aggregation. arXiv preprint arXiv:2401.13870(2024)

2024
[18]

Qiyao Ma, Xubin Ren, and Chao Huang. 2024. XRec: Large Language Models for Explainable Recommendation.arXiv preprint arXiv:2406.02377(2024)

2024 arXiv
[19]

Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton Van Den Hengel
[20]

Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme
[21]

Ilya Sutskever, James Martens, George Dahl, and Geoffrey Hinton. 2013. On the importance of initialization and momentum in deep learning. InInternational conference on machine learning. PMLR, 1139–1147

2013
[22]

Jinhui Tang, Xiaoyu Du, Xiangnan He, Fajie Yuan, Qi Tian, and Tat-Seng Chua
[23]

Zhulin Tao, Xiaohao Liu, Yewei Xia, Xiang Wang, Lifang Yang, Xianglin Huang, and Tat-Seng Chua. 2022. Self-supervised learning for multimedia recommenda- tion.IEEE Transactions on Multimedia(2022)

2022
[24]

Qifan Wang, Yinwei Wei, Jianhua Yin, Jianlong Wu, Xuemeng Song, and Liqiang Nie. 2021. Dualgnn: Dual graph neural network for multimedia recommendation. IEEE Transactions on Multimedia(2021)

2021
[25]

Wei Wei, Chao Huang, Lianghao Xia, and Chuxu Zhang. 2023. Multi-Modal Self-Supervised Learning for Recommendation. InProceedings of the ACM Web Conference 2023. 790–800

2023
[26]

Wei Wei, Xubin Ren, Jiabin Tang, Qinyong Wang, Lixin Su, Suqi Cheng, Jun- feng Wang, Dawei Yin, and Chao Huang. 2024. Llmrec: Large language models with graph augmentation for recommendation. InProceedings of the 17th ACM International Conference on Web Search and Data Mining. 806–815

2024
[27]

Yinwei Wei, Xiang Wang, Liqiang Nie, Xiangnan He, and Tat-Seng Chua. 2020. Graph-refined convolutional network for multimedia recommendation with implicit feedback. InProceedings of the 28th ACM international conference on multimedia. 3541–3549

2020
[28]

Yinwei Wei, Xiang Wang, Liqiang Nie, Xiangnan He, Richang Hong, and Tat-Seng Chua. 2019. MMGCN: Multi-modal graph convolution network for personalized recommendation of micro-video. InProceedings of the 27th ACM international conference on multimedia. 1437–1445

2019
[29]

Ashia C Wilson, Rebecca Roelofs, Mitchell Stern, Nati Srebro, and Benjamin Recht
[30]

Jinfeng Xu, Zheyu Chen, Jinze Li, Shuo Yang, Hewei Wang, and Edith CH Ngai
[31]

Jinfeng Xu, Zheyu Chen, Jinze Li, Shuo Yang, Wei Wang, Xiping Hu, and Edith C-H Ngai. 2024. FourierKAN-GCF: Fourier Kolmogorov-Arnold Network–An Effective and Efficient Feature Transformation for Graph Collaborative Filtering. arXiv preprint arXiv:2406.01034(2024)

2024 arXiv
[32]

Jinfeng Xu, Zheyu Chen, Wei Wang, Xiping Hu, Sang-Wook Kim, and Edith CH Ngai. 2025. COHESION: Composite Graph Convolutional Network with Dual- Stage Fusion for Multimodal Recommendation.arXiv preprint arXiv:2504.04452 (2025)

2025 arXiv
[33]

Jinfeng Xu, Zheyu Chen, Shuo Yang, Jinze Li, Hewei Wang, and Edith CH Ngai
[34]

Jinfeng Xu, Zheyu Chen, Shuo Yang, Jinze Li, Wei Wang, Xiping Hu, Steven Hoi, and Edith Ngai. 2025. A Survey on Multimodal Recommender Systems: Recent Advances and Future Directions.arXiv preprint arXiv:2502.15711(2025)

2025 arXiv
[35]

Guipeng Xv, Chen Lin, Wanxian Guan, Jinping Gou, Xubin Li, Hongbo Deng, Jian Xu, and Bo Zheng. 2023. E-commerce search via content collaborative graph neural network. InProceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining. 2885–2897

2023
[36]

InProceedings of the 33rd ACM International Conference on Information and Knowledge Management

AlignGroup: Learning and Aligning Group Consensus with Member Pref- erences for Group Recommendation. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. 2682–2691
[37]

Penghang Yu, Zhiyi Tan, Guanming Lu, and Bing-Kun Bao. 2023. Multi-view graph convolutional network for multimedia recommendation. InProceedings of the 31st ACM International Conference on Multimedia. 6576–6585

2023
[38]

Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. 2020. Gradient surgery for multi-task learning.Advances in Neural Information Processing Systems33 (2020), 5824–5836

2020
[39]

Jinghao Zhang, Yanqiao Zhu, Qiang Liu, Shu Wu, Shuhui Wang, and Liang Wang
[40]

Hongyu Zhou, Xin Zhou, Lingzi Zhang, and Zhiqi Shen. 2023. Enhancing dyadic relations with homogeneous graphs for multimodal recommendation. InECAI

2023
[41]

Xin Zhou. 2023. Mmrec: Simplifying multimodal recommendation. InProceedings of the 5th ACM International Conference on Multimedia in Asia Workshops. 1–2

2023
[42]

Xin Zhou. 2023. MMRec: Simplifying Multimodal Recommendation.arXiv preprint arXiv:2302.03497(2023)

2023 arXiv
[43]

Zhengyuan Yang, Linjie Li, Kevin Lin, Jianfeng Wang, Chung-Ching Lin, Zicheng Liu, and Lijuan Wang. 2023. The dawn of lmms: Preliminary explorations with gpt-4v (ision).arXiv preprint arXiv:2309.174219, 1 (2023), 1

2023 arXiv
[44]

Xin Zhou and Zhiqi Shen. 2023. A tale of two graphs: Freezing and denoising graph structures for multimodal recommendation. InProceedings of the 31st ACM International Conference on Multimedia. 935–943

2023
[45]

Xin Zhou, Hongyu Zhou, Yong Liu, Zhiwei Zeng, Chunyan Miao, Pengwei Wang, Yuan You, and Feijun Jiang. 2023. Bootstrap latent representations for multi-modal recommendation. InProceedings of the ACM Web Conference 2023. 845–854. MDVT KDD ’25, August 3–7, 2025, Toronto, ON, Cana...

2023
[52]

Xin Zhou and Chunyan Miao. 2024. Disentangled Graph Variational Auto- Encoder for Multimodal Recommendation With Interpretability.IEEE Transac- tions on Multimedia(2024)

2024
[2009]

InProceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence

BPR: Bayesian personalized ranking from implicit feedback. InProceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence. 452–461
[2015]

InProceedings of the 38th international ACM SIGIR conference on research and development in information retrieval

Image-based recommendations on styles and substitutes. InProceedings of the 38th international ACM SIGIR conference on research and development in information retrieval. 43–52
[2017]

Advances in neural information processing systems30 (2017)

The marginal value of adaptive gradient methods in machine learning. Advances in neural information processing systems30 (2017)

2017
[2019]

IEEE Transactions on Knowledge and Data Engineering32, 5 (2019), 855–867

Adversarial training towards robust multimedia recommender system. IEEE Transactions on Knowledge and Data Engineering32, 5 (2019), 855–867

2019
[2021]

InProceedings of the 29th ACM International Conference on Multimedia

Mining latent structures for multimedia recommendation. InProceedings of the 29th ACM International Conference on Multimedia. 3872–3880
[2023]

IOS Press, 3123–3130
[2024]

DiffMM: Multi-Modal Diffusion Model for Recommendation. (2024)

2024
[2025]

InProceedings of the AAAI Conference on Artificial Intelligence, Vol

Mentor: multi-level self-supervised learning for multimodal recommen- dation. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 12908–12917

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Léon Bottou, Frank E Curtis, and Jorge Nocedal. 2018. Optimization methods for large-scale machine learning.SIAM review60, 2 (2018), 223–311

2018

[2] [2]

Jiaben Chen, Xin Yan, Yihang Chen, Siyuan Cen, Qinwei Ma, Haoyu Zhen, Kaizhi Qian, Lie Lu, and Chuang Gan. 2024. RapVerse: Coherent Vocals and Whole-Body Motions Generations from Text.arXiv preprint arXiv:2405.20336(2024)

work page arXiv 2024

[3] [3]

Jingyuan Chen, Hanwang Zhang, Xiangnan He, Liqiang Nie, Wei Liu, and Tat- Seng Chua. 2017. Attentive collaborative filtering: Multimedia recommendation with item-and component-level attention. InProceedings of the 40th International ACM SIGIR conference on Research and Development in Information Retrieval. 335–344

work page 2017

[4] [4]

Xu Chen, Hanxiong Chen, Hongteng Xu, Yongfeng Zhang, Yixin Cao, Zheng Qin, and Hongyuan Zha. 2019. Personalized fashion recommendation with visual explanations based on multimodal attention network: Towards visually explainable recommendation. InProceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval...

2019

[5] [5]

Zheyu Chen, Jinfeng Xu, and Haibo Hu. 2025. Don’t Lose Yourself: Boosting Multimodal Recommendation via Reducing Node-neighbor Discrepancy in Graph Convolutional Network. InICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 1–5

work page 2025

[6] [6]

Zheyu Chen, Jinfeng Xu, Yutong Wei, and Ziyue Peng. 2025. Squeeze and Excita- tion: A Weighted Graph Contrastive Learning for Collaborative Filtering.arXiv preprint arXiv:2504.04443(2025)

arXiv 2025

[7] [7]

Zhiqiang Guo, Jianjun Li, Guohui Li, Chaoyang Wang, Si Shi, and Bin Ruan. 2024. LGMRec: Local and Global Graph Learning for Multimodal Recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 8454–8462

work page 2024

[8] [8]

Ruining He and Julian McAuley. 2016. VBPR: visual bayesian personalized ranking from implicit feedback. InProceedings of the AAAI conference on artificial intelligence, Vol. 30

2016