arxiv: 2604.26247 · v1 · submitted 2026-04-29 · 💻 cs.IR · cs.AI

Recognition: unknown

TimeMM: Time-as-Operator Spectral Filtering for Dynamic Multimodal Recommendation

Wei Yang , Rui Zhong , Zihan Lin , Xiaodan Wang , Cheng Chen , Huan Ren , Yao Hu

Authors on Pith no claims yet

Pith reviewed 2026-05-07 12:47 UTC · model grok-4.3

classification 💻 cs.IR cs.AI

keywords multimodal recommendationdynamic recommendationtemporal modelingspectral filteringuser preference evolutiongraph reweightingadaptive filtering

0 comments

The pith

TimeMM treats recency as a spectral operator on the user-item graph to adapt multimodal recommendations to non-stationary user preferences.

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

The paper sets out to show that user interests in multimodal recommendation systems evolve continuously and at different rates across visual and textual features, and that static graphs or simple time heuristics cannot capture this. It proposes modeling time itself as an operator that generates a bank of parametric kernels to reweight graph edges on the fly. These kernels are then mixed adaptively per prediction and per modality to produce effective spectral responses without computing full eigendecompositions. If the approach holds, recommendation models could track fine-grained preference shifts in linear time rather than relying on periodic retraining or coarse temporal bins.

Core claim

TimeMM instantiates Time-as-Operator by mapping interaction recency to a family of parametric temporal kernels that reweight edges on the user-item graph, producing component-specific representations without explicit eigendecomposition. Adaptive Spectral Filtering mixes the operator bank according to temporal context to yield prediction-specific effective spectral responses. Spectral-Aware Modality Routing calibrates visual and textual contributions under the same temporal context. Spectral Diversity Regularization encourages complementary behaviors across the filter bank and prevents collapse.

What carries the argument

Time-as-Operator, which maps recency to a family of parametric temporal kernels that reweight user-item graph edges to enable adaptive spectral filtering without eigendecomposition.

If this is right

The framework delivers linear-time scalability while outperforming existing multimodal recommenders on standard benchmarks.
Modality contributions are calibrated automatically according to temporal context rather than fixed weights.
Diversity regularization keeps the filter bank from collapsing to a single effective response.
Predictions become conditioned on continuous temporal regimes instead of discrete time slices.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The operator view of time could extend to other dynamic graph tasks such as session-based or sequential recommendation.
Production systems might reduce retraining frequency by relying on the online kernel mixing to track gradual preference drift.
A controlled simulation with synthetic non-stationary modality dominance would isolate whether the routing mechanism is the main driver of gains.

Load-bearing premise

That mapping recency to parametric temporal kernels and mixing them adaptively will capture non-stationary multimodal dynamics without post-hoc tuning or overfitting to specific datasets.

What would settle it

Run the model on a dataset containing documented abrupt shifts in visual versus textual preference dominance and measure whether the adaptive mixing produces measurably different spectral responses and higher ranking accuracy than fixed-kernel or non-temporal baselines.

Figures

Figures reproduced from arXiv: 2604.26247 by Cheng Chen, Huan Ren, Rui Zhong, Wei Yang, Xiaodan Wang, Yao Hu, Zihan Lin.

**Figure 1.** Figure 1: Overall architecture of TimeMM. TimeMM instantiates Time-as-Operator by mapping interaction recency to a family view at source ↗

**Figure 2.** Figure 2: Ablation study of TimeMM on three datasets. view at source ↗

**Figure 3.** Figure 3: Energy-decay signatures of TimeMM’s learned view at source ↗

**Figure 5.** Figure 5: Span-conditioned modality mixtures. Users are par view at source ↗

read the original abstract

Multimodal recommendation improves user modeling by integrating collaborative signals with heterogeneous item content. In real applications, user interests evolve over time and exhibit nonstationary dynamics, where different preference factors change at different rates. This challenge is amplified in multimodal settings because visual and textual cues can dominate decisions under different temporal regimes. Despite strong progress, most multimodal recommenders still rely on static interaction graphs or coarse temporal heuristics, which limits their ability to model continuous preference evolution with fine-grained temporal adaptation. To address these limitations, we propose TimeMM, a time-conditioned spectral filtering framework for dynamic multimodal recommendation. TimeMM instantiates Time-as-Operator by mapping interaction recency to a family of parametric temporal kernels that reweight edges on the user--item graph, producing component-specific representations without explicit eigendecomposition. To capture non-stationary interests, we introduce Adaptive Spectral Filtering that mixes the operator bank according to temporal context, yielding prediction-specific effective spectral responses. To account for modality-specific temporal sensitivity, we further propose Spectral-Aware Modality Routing that calibrates visual and textual contributions conditioned on the same temporal context. Finally, a ranking-space Spectral Diversity Regularization encourages complementary expert behaviors and prevents filter-bank collapse. Extensive experiments on real-world benchmarks demonstrate that TimeMM consistently outperforms state-of-the-art multimodal recommenders while maintaining linear-time scalability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

TimeMM frames time as an operator via parametric kernels and adaptive spectral mixing for dynamic multimodal recs, which is a concrete extension of prior spectral work, but the abstract gives no experimental numbers or ablations to back the outperformance claim.

read the letter

TimeMM maps recency to a bank of parametric temporal kernels that reweight user-item graph edges, then mixes those operators adaptively to produce context-specific spectral responses. It adds spectral-aware routing to let visual and textual signals contribute differently under different temporal regimes, plus a diversity regularizer to stop the bank from collapsing into one filter. These pieces are laid out clearly enough to see how they target non-stationary multimodal dynamics without explicit eigendecomposition, which is the main technical move that could support the linear-time claim if the mixing is done with cheap approximations like polynomial filters or message passing. The motivation about modalities dominating at different rates is practical and matches real systems. The construction itself is the part that stands out as new relative to static multimodal spectral recommenders. The soft spot is the missing evidence. The abstract states consistent gains over SOTA and linear scalability, yet supplies no baselines, metrics, error bars, ablation tables, or kernel parameterization details. Without those, it is impossible to tell whether the gains trace to the time-as-operator mechanism or simply to extra fitted capacity in the kernels and mixing coefficients. The stress-test worry about dataset-specific fitting rather than genuine generalization is worth checking once the full equations and regularization strength are visible. This is aimed at the multimodal recommendation community that already uses graph spectral methods and wants to add temporal adaptation. A reader already working on dynamic recsys would find the specific components worth examining. It deserves peer review because the framework is specific enough to test and the problem is real, even though the current writeup is light on results.

Referee Report

3 major / 1 minor

Summary. The paper proposes TimeMM, a time-conditioned spectral filtering framework for dynamic multimodal recommendation. It instantiates 'Time-as-Operator' by mapping interaction recency to a family of parametric temporal kernels that reweight edges on the user-item graph, introduces Adaptive Spectral Filtering to mix the operator bank according to temporal context for non-stationary interests, proposes Spectral-Aware Modality Routing to calibrate visual and textual contributions, and adds Spectral Diversity Regularization to encourage complementary behaviors. The central claim is that this yields consistent outperformance over state-of-the-art multimodal recommenders on real-world benchmarks while maintaining linear-time scalability without explicit eigendecomposition.

Significance. If the central claims hold, the work could meaningfully advance dynamic multimodal recommendation by providing a scalable spectral approach to modeling non-stationary, modality-specific preference evolution. The 'Time-as-Operator' framing and adaptive mixing of temporal kernels offer a principled alternative to static graphs or coarse heuristics, with potential to influence temporal GNN designs in recommendation. The linear scalability emphasis addresses a practical need for real-world deployment.

major comments (3)

[Abstract / Method] Abstract and method description: the claim that parametric temporal kernels and Adaptive Spectral Filtering produce effective, generalizable spectral responses for non-stationary dynamics is load-bearing, yet the exact kernel parameterization (reader notes parameter_count=2), the form of the mixing function, and any regularization beyond the diversity term are not shown to be independently derived or free of dataset-specific fitting; this leaves open the possibility that reported gains arise from added capacity rather than the proposed mechanism.
[Abstract] Abstract: the assertion of consistent outperformance and linear-time scalability supplies no experimental details, baselines, error bars, ablation results, or concrete implementation (e.g., polynomial filter or closed-form reweighting) that would verify O(|E|) scaling once the operator bank is instantiated; without these, the central empirical claim cannot be assessed.
[Method] Method: the Spectral-Aware Modality Routing and operator-bank mixing are described at a high level, but no analysis demonstrates that the adaptive responses track modality-specific temporal regimes without overfitting to the training distribution or requiring post-hoc tuning, which directly affects whether the non-stationary modeling claim holds.

minor comments (1)

[Abstract] The abstract could more explicitly name the benchmarks and metrics used to support the outperformance claim.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below, providing clarifications from the full paper and committing to targeted revisions that strengthen the presentation of our claims.

read point-by-point responses

Referee: [Abstract / Method] Abstract and method description: the claim that parametric temporal kernels and Adaptive Spectral Filtering produce effective, generalizable spectral responses for non-stationary dynamics is load-bearing, yet the exact kernel parameterization (reader notes parameter_count=2), the form of the mixing function, and any regularization beyond the diversity term are not shown to be independently derived or free of dataset-specific fitting; this leaves open the possibility that reported gains arise from added capacity rather than the proposed mechanism.

Authors: We acknowledge the need for greater explicitness here. The manuscript defines the temporal kernels in Equation (3) as a two-parameter family (decay rate and scaling factor) chosen to flexibly approximate recency-based edge reweighting while remaining differentiable and closed-form. The mixing function is a softmax over temporal-context embeddings applied to the operator bank, as derived in Section 3.2 from the requirement for prediction-specific spectral responses. The diversity term is the primary regularizer, but the overall design is grounded in spectral graph theory to promote generalizability rather than capacity alone. To directly address the concern, we will expand the method section with a derivation subsection and add synthetic-data experiments isolating the mechanism from capacity effects in the revision. revision: yes
Referee: [Abstract] Abstract: the assertion of consistent outperformance and linear-time scalability supplies no experimental details, baselines, error bars, ablation results, or concrete implementation (e.g., polynomial filter or closed-form reweighting) that would verify O(|E|) scaling once the operator bank is instantiated; without these, the central empirical claim cannot be assessed.

Authors: The abstract serves as a concise summary; the supporting details appear in Sections 4 and 5, which report results against multimodal baselines (MMGCN, LightGCN, temporal variants), mean and standard deviation over five runs, ablation tables, and complexity analysis confirming O(|E|) scaling via closed-form kernel reweighting without eigendecomposition. We will revise the abstract to include a brief clause referencing the experimental protocol and complexity result, while ensuring the claims remain fully substantiated by the detailed sections. revision: partial
Referee: [Method] Method: the Spectral-Aware Modality Routing and operator-bank mixing are described at a high level, but no analysis demonstrates that the adaptive responses track modality-specific temporal regimes without overfitting to the training distribution or requiring post-hoc tuning, which directly affects whether the non-stationary modeling claim holds.

Authors: We agree that explicit validation of adaptation behavior would strengthen the non-stationary claim. The routing mechanism conditions modality weights on the same temporal embedding used for operator mixing (Section 3.3), and the paper already reports standard hyperparameter search without post-hoc tuning. In the revision we will add (i) visualizations of routing weights across temporal buckets and (ii) results on temporally held-out splits to confirm tracking of modality-specific regimes without overfitting. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation chain is self-contained with empirical claims

full rationale

The provided abstract and description outline a time-conditioned spectral filtering approach using parametric temporal kernels, adaptive mixing, modality routing, and a diversity regularizer. No equations, derivations, or self-citations are present that reduce any claimed prediction or result to its inputs by construction. Performance claims are framed as empirical outcomes on benchmarks rather than analytic identities. The method's components are described as novel constructions without invoking uniqueness theorems or prior self-work to force the form. This is the common case of a non-circular proposal whose validity rests on external validation rather than internal tautology.

Axiom & Free-Parameter Ledger

2 free parameters · 0 axioms · 1 invented entities

The framework rests on several parametric components whose values are not derived from first principles in the abstract.

free parameters (2)

parametric temporal kernels
Family of kernels that reweight user-item edges based on interaction recency.
operator bank mixing coefficients
Parameters that blend spectral responses according to temporal context.

invented entities (1)

Time-as-Operator no independent evidence
purpose: Mapping recency to temporal kernels for graph reweighting
New conceptual operator introduced to handle continuous time without explicit eigendecomposition.

pith-pipeline@v0.9.0 · 5545 in / 1129 out tokens · 55080 ms · 2026-05-07T12:47:35.298478+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

108 extracted references · 26 canonical work pages · 6 internal anchors

[1]

Keqin Bao, Jizhi Zhang, Wenjie Wang, Yang Zhang, Zhengyi Yang, Yanchen Luo, Chong Chen, Fuli Feng, and Qi Tian. 2025. A bi-step grounding paradigm for large language models in recommendation systems.ACM Transactions on Recommender Systems3, 4 (2025), 1–27

2025
[2]

Keqin Bao, Jizhi Zhang, Yang Zhang, Wenjie Wang, Fuli Feng, and Xiangnan He. 2023. Tallrec: An effective and efficient tuning framework to align large language model with recommendation. InProceedings of the 17th ACM conference on recommender systems. 1007–1014

2023
[3]

Shuqing Bian, Xingyu Pan, Wayne Xin Zhao, Jinpeng Wang, Chuyuan Wang, and Ji-Rong Wen. 2023. Multi-modal mixture of experts represetation learning for sequential recommendation. InProceedings of the 32nd ACM international conference on information and knowledge management. 110–119

2023
[4]

Ching Chang, Yidan Shi, Defu Cao, Wei Yang, Jeehyun Hwang, Haixin Wang, Jiacheng Pang, Wei Wang, Yan Liu, Wen-Chih Peng, et al . 2025. A survey of reasoning and agentic systems in time series with large language models.arXiv preprint arXiv:2509.11575(2025)

work page arXiv 2025
[5]

Lihua Chen, Ning Yang, and Philip S Yu. 2022. Time lag aware sequential recom- mendation. InProceedings of the 31st ACM international conference on information & knowledge management. 212–221

2022
[6]

Weixin Chen, Yuhan Zhao, Jingyuan Huang, Zihe Ye, Clark Mingxuan Ju, Tong Zhao, Neil Shah, Li Chen, and Yongfeng Zhang. 2026. MemRec: Col- laborative Memory-Augmented Agentic Recommender System.arXiv preprint arXiv:2601.08816(2026)

work page internal anchor Pith review Pith/arXiv arXiv 2026
[7]

Yiqun Chen, Jinyuan Feng, Wei Yang, Meizhi Zhong, Zhengliang Shi, Rui Li, Xiaochi Wei, Yan Gao, Yi Wu, Yao Hu, et al. 2026. Self-Compression of Chain-of- Thought via Multi-Agent Reinforcement Learning.arXiv preprint arXiv:2601.21919 (2026)

work page arXiv 2026
[8]

Yiqun Chen, Qi Liu, Yi Zhang, Weiwei Sun, Xinyu Ma, Wei Yang, Daiting Shi, Jiaxin Mao, and Dawei Yin. 2025. Tourrank: Utilizing large language models for documents ranking with a tournament-inspired strategy. InProceedings of the ACM on Web Conference 2025. 1638–1652

2025
[9]

Yashar Deldjoo, Markus Schedl, and Peter Knees. 2021. Content-driven music recommendation: Evolution, state of the art, and challenges.arXiv preprint arXiv:2107.11803(2021)

work page arXiv 2021
[10]

Xinyu Du, Huanhuan Yuan, Pengpeng Zhao, Jianfeng Qu, Fuzhen Zhuang, Guan- feng Liu, Yanchi Liu, and Victor S Sheng. 2023. Frequency enhanced hybrid attention network for sequential recommendation. InProceedings of the 46th International ACM SIGIR conference on research and development in information retrieval. 78–88

2023
[11]

Ziwei Fan, Zhiwei Liu, Jiawei Zhang, Yun Xiong, Lei Zheng, and Philip S Yu
[12]

InProceedings of the 30th ACM international conference on information & knowledge management

Continuous-time sequential recommendation with temporal graph collab- orative transformer. InProceedings of the 30th ACM international conference on information & knowledge management. 433–442
[13]

Zhenyi Fan, Hongbin Zhang, Guangyu Lin, Lianglun Cheng, Zhuowei Wang, and Chong Chen. 2024. Hierarchical Multi-Frequency Transform for Sequential Recommendation. In2024 27th International Conference on Computer Supported Cooperative Work in Design (CSCWD). IEEE, 1559–1564

2024
[14]

Hao Gu, Rui Zhong, Yu Xia, Wei Yang, Chi Lu, Peng Jiang, and Kun Gai. 2025. R4ec: A Reasoning, Reflection, and Refinement Framework for Recommendation Systems.arXiv preprint arXiv:2507.17249(2025)

work page arXiv 2025
[15]

Tengyue Han, Pengfei Wang, Shaozhang Niu, and Chenliang Li. 2022. Modality matches modality: Pretraining modality-disentangled item representations for recommendation. InProceedings of the ACM Web Conference 2022. 2058–2066

2022
[16]

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
[17]

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
[18]

Nana Huang, Ruimin Hu, Xiaochen Wang, and Hongwei Ding. 2023. Multi- scale modeling temporal hierarchical attention for sequential recommendation. Information Sciences641 (2023), 119126

2023
[19]

Nana Huang, Ruimin Hu, Mingfu Xiong, Xiaoran Peng, Hongwei Ding, Xiaodong Jia, and Lingkun Zhang. 2022. Multi-scale interest dynamic hierarchical trans- former for sequential recommendation.Neural Computing and Applications34, 19 (2022), 16643–16654

2022
[20]

Yuanming Huang, Jie Lu, Keqiuyin Li, and Guangquan Zhang. 2025. Learning a Wavelet Neural Filter with Mamba for Sequential Recommendation. In2025 IEEE Symposium on Computational Intelligence in Image, Signal Processing and Synthetic Media (CISM). IEEE, 1–7

2025
[21]

Hao Jiang, Wenjie Wang, Yinwei Wei, Zan Gao, Yinglong Wang, and Liqiang Nie
[22]

InProceedings of the 28th ACM International conference on Multimedia

What aspect do you like: Multi-scale time-aware user interest modeling for micro-video recommendation. InProceedings of the 28th ACM International conference on Multimedia. 3487–3495
[23]

Yangqin Jiang, Lianghao Xia, Wei Wei, Da Luo, Kangyi Lin, and Chao Huang
[24]

InProceedings of the 32nd ACM International Conference on Multimedia

Diffmm: Multi-modal diffusion model for recommendation. InProceedings of the 32nd ACM International Conference on Multimedia. 7591–7599
[25]

Bryce Kan, Wei Yang, Emily Nguyen, Ganghui Yi, Bowen Yi, Chenxiao Yu, and Yan Liu. 2026. De-conflating Preference and Qualification: Constrained Dual- Perspective Reasoning for Job Recommendation with Large Language Models. arXiv preprint arXiv:2602.03097(2026). TimeMM: Time-as-Operator Spectral Filtering for Dynamic Multimodal Recommendation SIGIR ’26, J...

work page arXiv 2026
[26]

Wang-Cheng Kang, Chen Fang, Zhaowen Wang, and Julian McAuley. 2017. Visually-aware fashion recommendation and design with generative image mod- els. In2017 IEEE international conference on data mining (ICDM). IEEE, 207–216

2017
[27]

Saketh Reddy Karra and Theja Tulabandhula. 2024. Interarec: Interactive recom- mendations using multimodal large language models. InPacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, 32–43

2024
[28]

Chanwoo Kim, Jinkyu Sung, Yebonn Han, and Joonseok Lee. 2025. Graph Spectral Filtering with Chebyshev Interpolation for Recommendation.arXiv preprint arXiv:2505.00552(2025)

work page arXiv 2025
[29]

Hye-young Kim, Minjin Choi, Sunkyung Lee, Ilwoong Baek, and Jongwuk Lee
[30]

InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval

DIFF: Dual Side-Information Filtering and Fusion for Sequential Recom- mendation. InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1624–1633
[31]

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

work page internal anchor Pith review arXiv 2014
[32]

Yehuda Koren. 2009. Collaborative filtering with temporal dynamics. InProceed- ings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining. 447–456

2009
[33]

Honghao Li, Lei Sang, Yi Zhang, and Yiwen Zhang. 2024. SimCEN: Simple Contrast-enhanced Network for CTR Prediction. InProceedings of the 32nd ACM International Conference on Multimedia. 2311–2320

2024
[34]

Honghao Li, Yiwen Zhang, Yi Zhang, Lei Sang, and Jieming Zhu. 2025. Revisiting Feature Interactions from the Perspective of Quadratic Neural Networks for Click-through Rate Prediction. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2. 1365–1375

2025
[35]

Jiacheng Li, Yujie Wang, and Julian McAuley. 2020. Time interval aware self- attention for sequential recommendation. InProceedings of the 13th international conference on web search and data mining. 322–330

2020
[36]

Li Li, Peilin Cai, Ryan A Rossi, Franck Dernoncourt, Branislav Kveton, Junda Wu, Tong Yu, Linxin Song, Tiankai Yang, Yuehan Qin, et al. 2025. A personalized con- versational benchmark: Towards simulating personalized conversations.arXiv preprint arXiv:2505.14106(2025)

work page arXiv 2025
[37]

Li Li, Wei Ji, Yiming Wu, Mengze Li, You Qin, Lina Wei, and Roger Zimmermann
[38]

In Proceedings of the AAAI conference on artificial intelligence, Vol

Panoptic scene graph generation with semantics-prototype learning. In Proceedings of the AAAI conference on artificial intelligence, Vol. 38. 3145–3153
[39]

Li Li, You Qin, Wei Ji, Yuxiao Zhou, and Roger Zimmermann. 2024. Domain-wise invariant learning for panoptic scene graph generation. InICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 3165–3169

2024
[40]

Li Li, Chenwei Wang, You Qin, Wei Ji, and Renjie Liang. 2023. Biased-predicate annotation identification via unbiased visual predicate representation. InPro- ceedings of the 31st ACM International Conference on Multimedia. 4410–4420

2023
[41]

Shawn Li, Huixian Gong, Hao Dong, Tiankai Yang, Zhengzhong Tu, and Yue Zhao
[42]

InProceedings of the Computer Vision and Pattern Recognition Conference

Dpu: Dynamic prototype updating for multimodal out-of-distribution de- tection. InProceedings of the Computer Vision and Pattern Recognition Conference. 10193–10202
[43]

Shixuan Li, Wei Yang, Peiyu Zhang, Xiongye Xiao, Defu Cao, Yuehan Qin, Xiaole Zhang, Yue Zhao, and Paul Bogdan. 2025. Climatellm: Efficient weather forecast- ing via frequency-aware large language models.arXiv preprint arXiv:2502.11059 (2025)

work page arXiv 2025
[44]

Yuecheng Li, Hengwei Ju, Zeyu Song, Wei Yang, Chi Lu, Peng Jiang, and Kun Gai. 2026. RecGOAT: Graph Optimal Adaptive Transport for LLM-Enhanced Multimodal Recommendation with Dual Semantic Alignment.arXiv preprint arXiv:2602.00682(2026)

work page arXiv 2026
[45]

Han Liu, Yinwei Wei, Fan Liu, Wenjie Wang, Liqiang Nie, and Tat-Seng Chua
[46]

Dynamic multimodal fusion via meta-learning towards micro-video rec- ommendation.ACM Transactions on Information Systems42, 2 (2023), 1–26

2023
[47]

Jiahao Liu, Shengkang Gu, Dongsheng Li, Guangping Zhang, Mingzhe Han, Hansu Gu, Peng Zhang, Tun Lu, Li Shang, and Ning Gu. 2025. AgentCF++: Memory-enhanced LLM-based Agents for Popularity-aware Cross-domain Rec- ommendations. InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2566–2571

2025
[48]

Yuqing Liu, Yu Wang, Lichao Sun, and Philip S Yu. 2024. Rec-gpt4v: Mul- timodal recommendation with large vision-language models.arXiv preprint arXiv:2402.08670(2024)

work page arXiv 2024
[49]

Yuxi Liu, Lianghao Xia, and Chao Huang. 2024. Selfgnn: Self-supervised graph neural networks for sequential recommendation. InProceedings of the 47th In- ternational ACM SIGIR Conference on Research and Development in Information Retrieval. 1609–1618

2024
[50]

Yifan Liu, Kangning Zhang, Xiangyuan Ren, Yanhua Huang, Jiarui Jin, Yingjie Qin, Ruilong Su, Ruiwen Xu, Yong Yu, and Weinan Zhang. 2024. AlignRec: Aligning and Training in Multimodal Recommendations. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. 1503–1512

2024
[51]

Yusheng Lu and Yongrui Duan. 2024. Online content-based sequential rec- ommendation considering multimodal contrastive representation and dynamic preferences.Neural Computing and Applications36, 13 (2024), 7085–7103

2024
[52]

Haitong Luo, Xuying Meng, Suhang Wang, Hanyun Cao, Weiyao Zhang, Yequan Wang, and Yujun Zhang. 2024. Spectral-based graph neural networks for com- plementary item recommendation. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 8868–8876

2024
[53]

Shijie Luo, Jianxia Chen, Tianci Yu, Shi Dong, Gaohang Jiang, and Ninglong Ding. 2024. Dual Frequency-based Temporal Sequential Recommendation. In 2024 International Joint Conference on Neural Networks (IJCNN). IEEE, 1–8

2024
[54]

Minh-Duc Nguyen, Hai-Dang Kieu, and Dung D Le. 2026. AMEM4Rec: Leveraging Cross-User Similarity for Memory Evolution in Agentic LLM Recommenders. arXiv preprint arXiv:2602.08837(2026)

work page arXiv 2026
[55]

Rongqing Kenneth Ong and Andy WH Khong. 2025. Spectrum-based modality representation fusion graph convolutional network for multimodal recommen- dation. InProceedings of the Eighteenth ACM International Conference on Web Search and Data Mining. 773–781

2025
[56]

Heng Ping, Arijit Bhattacharjee, Peiyu Zhang, Shixuan Li, Wei Yang, Anzhe Cheng, Xiaole Zhang, Jesse Thomason, Ali Jannesari, Nesreen Ahmed, et al
[57]

Verimoa: A mixture-of-agents framework for spec-to-hdl generation.arXiv preprint arXiv:2510.27617(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[58]

Heng Ping, Shixuan Li, Peiyu Zhang, Anzhe Cheng, Shukai Duan, Nikos Kanakaris, Xiongye Xiao, Wei Yang, Shahin Nazarian, Andrei Irimia, et al
[59]

In2025 IEEE International Conference on LLM-Aided Design (ICLAD)

Hdlcore: A training-free framework for mitigating hallucinations in llm- generated hdl. In2025 IEEE International Conference on LLM-Aided Design (ICLAD). IEEE, 108–116
[60]

Heng Ping, Peiyu Zhang, Zhenkun Wang, Shixuan Li, Anzhe Cheng, Wei Yang, Paul Bogdan, and Shahin Nazarian. 2026. POET: Power-Oriented Evolutionary Tuning for LLM-Based RTL PPA Optimization.arXiv preprint arXiv:2603.19333 (2026)

work page arXiv 2026
[61]

Yifang Qin, Wei Ju, Hongjun Wu, Xiao Luo, and Ming Zhang. 2024. Learning graph ode for continuous-time sequential recommendation.IEEE Transactions on Knowledge and Data Engineering36, 7 (2024), 3224–3236

2024
[62]

Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme
[63]

BPR: Bayesian personalized ranking from implicit feedback.arXiv preprint arXiv:1205.2618(2012)

work page internal anchor Pith review arXiv 2012
[64]

Marian-Andrei Rizoiu, Lexing Xie, Scott Sanner, Manuel Cebrian, Honglin Yu, and Pascal Van Hentenryck. 2017. Expecting to be hip: Hawkes intensity processes for social media popularity. InProceedings of the 26th international conference on world wide web. 735–744

2017
[65]

Juntao Tan, Shuyuan Xu, Wenyue Hua, Yingqiang Ge, Zelong Li, and Yongfeng Zhang. 2024. Idgenrec: Llm-recsys alignment with textual id learning. InProceed- ings of the 47th international ACM SIGIR conference on research and development in information retrieval. 355–364

2024
[66]

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
[67]

Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971(2023)

work page internal anchor Pith review arXiv 2023
[68]

Ty Valencia, Burak Barlas, Varun Singhal, Ruchir Bhatia, and Wei Yang. 2026. VLM4Rec: Multimodal Semantic Representation for Recommendation with Large Vision-Language Models.arXiv preprint arXiv:2603.12625(2026)

work page arXiv 2026
[69]

Kunal Vaswani, Yudhik Agrawal, and Vinoo Alluri. 2021. Multimodal fusion based attentive networks for sequential music recommendation. In2021 IEEE seventh international conference on multimedia big data (BigMM). IEEE, 25–32

2021
[70]

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 Multimedia25 (2021), 1074–1084

2021
[71]

Wei Wei, Jiabin Tang, Lianghao Xia, Yangqin Jiang, and Chao Huang. 2024. Promptmm: Multi-modal knowledge distillation for recommendation with prompt-tuning. InProceedings of the ACM Web Conference 2024. 3217–3228

2024
[72]

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
[73]

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
[74]

Fang Wu and Bernardo A Huberman. 2007. Novelty and collective attention. Proceedings of the National Academy of Sciences104, 45 (2007), 17599–17601

2007
[75]

Zihao Wu, Xin Wang, Heng Chang, Hong Chen, Lifeng Sun, and Wenwu Zhu
[76]

InProceedings of the 2025 International Conference on Multimedia Retrieval

Aligning Large Multimodal Model with Sequential Recommendation via Content-Behavior Guidance. InProceedings of the 2025 International Conference on Multimedia Retrieval. 1507–1516

2025
[77]

Lianghao Xia, Chao Huang, Yong Xu, and Jian Pei. 2022. Multi-behavior sequen- tial recommendation with temporal graph transformer.IEEE Transactions on Knowledge and Data Engineering35, 6 (2022), 6099–6112. SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia Wei Yang et al

2022
[78]

Yu Xia, Rui Zhong, Hao Gu, Wei Yang, Chi Lu, Peng Jiang, and Kun Gai. 2025. Hierarchical Tree Search-based User Lifelong Behavior Modeling on Large Lan- guage Model. InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1758–1767

2025
[79]

Yu Xia, Rui Zhong, Zeyu Song, Wei Yang, Junchen Wan, Qingpeng Cai, Chi Lu, and Peng Jiang. 2025. TrackRec: Iterative Alternating Feedback with Chain- of-Thought via Preference Alignment for Recommendation.arXiv preprint arXiv:2508.15388(2025)

work page arXiv 2025
[80]

Jingyu Xu, Bo Yang, Zimu Li, Wei Liu, and Hao Qiao. 2025. FAGCL: frequency- based augmentation graph contrastive learning for recommendation.Applied Intelligence55, 1 (2025), 44

2025

Showing first 80 references.