Pith. sign in

REVIEW 3 major objections 4 minor 37 references

A recommender that uses a Kalman filter's uncertainty to blend static and temporal evidence reports the strongest cold-start results across two datasets and twelve baselines.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 15:58 UTC pith:YRR2BJVU

load-bearing objection A well-built fusion architecture whose central time-awareness claim is undercut by its own diagnostic figure; worth referee time but needs artifact release and reconciliation. the 3 major comments →

arxiv 2607.26429 v1 pith:YRR2BJVU submitted 2026-07-29 cs.IR

NMKFR: A Robust Framework for Time-Aware Cold-Start Recommendation

Chengzhi Liu , Ning Zeng , Zehui Qu This is my paper
classification cs.IR
keywords item cold-start recommendationtime-aware recommendationKalman filteringposterior covarianceuncertainty-guided fusionsemantic encodingTitans long-term memorysampled candidate ranking
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Cold-start recommendation is hard not only because new items have few interactions, but because the reliability of the evidence—item text, early feedback, and temporal context—changes over the item's lifecycle. This paper proposes NMKFR, a two-branch recommender that encodes item text as a semantic observation and feeds it into a time-aware Kalman filter tracking a latent temporal state. The trace of the filter's posterior covariance becomes an uncertainty signal that decides how much to trust static semantics versus dynamic state evidence, and also modulates memory retrieval in the text encoder. On Amazon Video Games and MovieLens-32M, under time-aware and item cold-start protocols, NMKFR reports the strongest ranking results among twelve baselines, with ablations and robustness analyses supporting the uncertainty-guided fusion. If correct, the paper establishes that a principled uncertainty estimate—not just more or stronger signals—is a workable lever for cold-start ranking.

Core claim

The paper's central claim is that posterior-covariance-guided semantic–temporal fusion improves time-aware item cold-start recommendation. Concretely, the development is a dual-branch architecture: FD-TSE uses a Titans-style memory-augmented encoder to turn item text into a static semantic vector; TA-KDT projects that vector into an observation of a user-conditioned latent state and applies a Kalman prediction–update with a time-gap-dependent transition; and UDFM/ACFM use the posterior covariance trace Ut = tr(Pt|t) to bias both memory retrieval and the softmax fusion of static and dynamic features. The reported experiments (Table 1) show NMKFR with the best overall performance under both pr

What carries the argument

The load-bearing object is the posterior covariance matrix of the Kalman filter, summarized by its trace Ut = tr(Pt|t). This one scalar is used twice: the current trace biases the softmax weights in the adaptive comparison fusion module (ACFM), shifting the balance between the static semantic branch and the dynamic temporal branch, while the preceding trace modulates FD-TSE's memory retrieval by interpolating between stored memory and the current token embedding. The design encodes the paper's central intuition that uncertainty—not just the point estimate of the state—should drive evidence weighting.

Load-bearing premise

The framework's temporal branch assumes the user-conditioned latent state evolves as a linear Gaussian process (with log-compressed time gaps in the transition), and that the trace of the posterior covariance is a faithful single-number proxy for how much to trust the temporal estimate; if real state dynamics are strongly nonlinear, or that trace is not informative about actual state error, the uncertainty-guided fusion has no principled advantage.

What would settle it

A controlled comparison that replaces the posterior-trace signal with a fixed, untuned fusion weight while keeping all other modules identical, and finds no reliable ranking difference on either dataset, would falsify the claim that the uncertainty signal is what drives the gains. A sharper probe: in a synthetic environment with a known latent state, if the posterior trace does not rise when the temporal-state error actually grows, the signal is not calibrated.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Reported gains: under time-aware cold-start, NMKFR improves over the strongest baseline by at least 2.16% (Amazon Video Games) and 3.39% (MovieLens-32M); under item cold-start, by at least 3.10% and 0.40%, with stronger MovieLens gains on ranking-sensitive metrics like NDCG@10 and MRR.
  • Ablation study: removing any of the five components (Kalman tracking, Titans encoder, UDFM, ACFM, explicit Δt modeling) hurts performance; removing ACFM causes the largest NDCG@10 drop on Amazon Video Games, supporting comparison-based fusion.
  • Adaptive posterior-guided fusion outperforms fixed-gain fusion under token-level input noise up to 0.20 and across limited-history prefixes from 0 to 10 user interactions.
  • Kalman diagnostics: the posterior trace and innovation-scale proxy remain bounded during optimization and across synthetic temporal gaps up to 1095 days, and bucket-level NDCG peaks in the 7–30 day temporal-gap interval.
  • If the central claim holds, uncertainty-guided fusion is a portable principle for any cold-start setting where evidence sources have time-varying reliability.

Where Pith is reading between the lines

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

  • Editorial inference: the covariance-trace control is a generic uncertainty-fusion primitive; it could be ported to any two-modality ranker where one modality's trustworthiness decays with time, such as image-plus-text cold start or news recommendation with breaking-item volatility.
  • Editorial inference: the paper leaves untested whether the posterior trace is the best uncertainty summary; comparing against the maximum eigenvalue or full-covariance norm of Pt|t on the same control stream would isolate the trace's specific role.
  • Editorial inference: because the noise matrices Qt and Rt are defined only up to a parameterization 'given in the supplementary material' that is not in this preprint, a direct reimplementation would need to treat that parameterization as a free design choice, which could affect the reported margins.
  • Editorial inference: a synthetic environment with known state drift could directly test whether the posterior trace is calibrated to true state error, separating the uncertainty signal's real effect from mere correlation with ranking difficulty.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper presents NMKFR, a dual-branch recommender for time-aware item cold-start. FD-TSE encodes item text with a Titans-style memory-augmented transformer; TA-KDT tracks a user-conditioned latent state via a Kalman filter with a time-dependent transition; UDFM converts the posterior covariance trace into an uncertainty signal; and ACFM uses that signal to fuse static and dynamic features before ranking. Experiments on Amazon Video Games and MovieLens-32M under time-aware and item cold-start protocols report best overall results over 12 baselines, with ablations, posterior diagnostics, and robustness analyses.

Significance. If the reported gains hold, NMKFR would be a useful demonstration of using a principled state-space uncertainty measure to control static-temporal fusion in cold-start recommendation. The paper's breadth of baselines and its explicit diagnostic design are strengths: the authors attempt to inspect the mechanism rather than only report aggregate metrics. However, the manuscript currently does not support its strongest claims in a reproducible way. The ablation stream in Table 2 yields different absolute numbers from the main protocol in Table 1; no variance is reported; the Kalman noise parameterization is in a missing appendix; and the synthetic-gap diagnostic in Figure 6 indicates that the temporal branch is nearly insensitive to the time gaps it is designed to model. These are load-bearing correctness concerns, not presentation issues.

major comments (3)
  1. [Section 4, Table 2 vs Table 1] The Full NMKFR row in Table 2 reports MovieLens R@10=0.5822 and N@10=0.4066, while Table 1 reports R@10=0.4966 (time-aware) and 0.5393 (item cold-start), and N@10=0.3389/0.3109. Amazon values differ similarly (R@10 0.4410 vs 0.4256/0.3823). The text says the ablation uses 'the same split files, 101-candidate construction, and evaluation metrics' within a controlled stream, but it does not explain how this stream differs from the main protocol. Since RQ2 uses Table 2 to attribute gains to components, this unexplained discrepancy prevents the reader from validating the ablation-based conclusions. Please reconcile the two evaluation streams, e.g., identify the sampled subset or checkpoint used in Table 2, or report Table 1 numbers in the ablation table.
  2. [Section 3, Eq. (9) and Section 4, Fig. 6] Eq. (9) defines Q_t and R_t through 'bounded positive scales' whose parameterization is deferred to a supplementary appendix that is not included in the preprint. Independently of that, Figure 6 shows that over input gaps from 0 to 1095 days, the innovation proxy, posterior trace, and Kalman-gain norm change by at most 0.021%, 0.005%, and 0.005% on Amazon Video Games, and 0.20%, 0.037%, and 0.035% on MovieLens-32M. Eq. (16) uses log(tr(P_{t|t})+epsilon) as the fusion bias, so a near-constant posterior trace implies the uncertainty signal carries essentially no temporal-gap information. The paper describes this as bounded numerical stability, but the stated contribution is 'time-aware Kalman state tracking' and 'posterior-covariance-guided semantic-temporal fusion.' A branch that is invariant to the time gap cannot support the time-aware mechanism central to the paper. Please either show
  3. [Section 4, Evaluation Metrics] The paper says results are averaged over three runs, but Tables 1 and 2 and all figures report point estimates without standard deviations or confidence intervals. Several key margins in Table 1 are small on ranking-sensitive metrics (e.g., MovieLens time-aware R@10: 0.4966 vs SASRec 0.4759), so without variance information the reader cannot assess whether the stated SOTA improvements are statistically meaningful. Please report per-seed results or standard deviations, and ideally a significance test for the main Table 1 comparisons.
minor comments (4)
  1. [Section 3, Eq. (7)] The Normalize operation in Eq. (7) is not defined. Please specify the normalization (e.g., min-max over training timestamps) and indicate whether it is fit on the training set.
  2. [Section 3, Eq. (4)] The text says exp(-clamp(\bar U_t, 10)) is used for numerical stability, while Eq. (4) shows e^{-\bar U_t}. Please reconcile the notation and clarify whether the clamping is applied inside the exponent.
  3. [Figure 6 caption] The caption states changes are 'relative to the zero-day reference,' but it is not clear whether the reference is the first plotted point or the training-distribution average. Please define the reference explicitly so the percentages are interpretable.
  4. [Supplementary material references] Several passages refer to supplementary material for dataset statistics, seed details, noise scale parameterization, and a timestamped semantic-evidence audit. The supplementary material is not included with the preprint; please provide it or state how it can be obtained.

Circularity Check

0 steps flagged

No significant circularity: NMKFR's components are trained and evaluated on external benchmarks; no fitted value is renamed as a prediction and no load-bearing self-citation is present.

full rationale

This is an empirical ML paper rather than a formal derivation chain. The claimed result is ranking effectiveness on held-out test streams under fixed splits and candidate sets. The core components (FD-TSE, TA-KDT, UDFM, ACFM) are defined as model architecture, and the posterior covariance trace in Eq. 13 is computed from the Kalman recursions in Eqs. 10-11 using observed semantic vectors; it is not fitted to the ranking metrics that the paper claims to predict. The UDFM fusion weight in Eq. 16 is an architectural use of that computed signal, and its contribution is tested by independently retrained ablations and fixed-gain comparisons, not by re-predicting the same fitted values. The paper does not invoke a uniqueness theorem, does not rely on load-bearing self-citations (the related-work citations are external baselines or standard prior methods), and does not rename a known empirical result as a new derivation. There are legitimate non-circular concerns: Figure 6 shows that the Kalman quantities move by less than 0.04% over 0-1095 day gaps, which suggests the temporal branch may be nearly time-invariant and undercuts the central 'time-aware' claim, and Table 2's full NMKFR row differs substantially from Table 1 despite the text describing the same split files and 101-candidate construction. However, these are correctness, consistency, or effectiveness issues, not circularity: neither concern reduces a claimed output to its own inputs by construction. Accordingly, the honest finding is no significant circularity.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 1 invented entities

The framework rests on the standard ML training pipeline plus several design assumptions. The most load-bearing are the linear-Gaussian Kalman model and the scalar trace uncertainty signal; both are design choices with only internal diagnostic support. The missing supplementary blocks auditing of the noise parameterization. No new physical entities are introduced; the latent temporal state is a model variable.

free parameters (6)
  • State dimension k = 32
    Chosen for the Kalman latent state (Section 3, TA-KDT). No sensitivity analysis or justification provided. Affects expressiveness and the scale of the posterior trace.
  • Clamping bound for uncertainty signal = 10
    Eq. (4) uses exp(-clamp(U_t, 10)); the clamp upper bound is chosen by hand and controls memory interpolation.
  • Learned fusion bias coefficient λ = learned
    Eq. (16) uses λ log(U_t+ε) as a bias before softmax; λ is a learned scalar fitted during training.
  • Kalman process/observation noise scales q0, r0
    Eq. (9) defines Q_t = q0 s_t^Q I_k and R_t = r0 s_t^R I_k; the parameterization of s_t^Q and s_t^R is deferred to supplementary material, so the effective noise parameters are not specified in the paper.
  • Loss weights λ1, λ2, λ3
    Eq. (19) linearly combines ListNet, pairwise, and BCE losses; the weights are not reported, so the relative supervision strength is unspecified.
  • Time compression Normalize/Log1p
    Eq. (7) compresses time intervals; the normalization constants in 'Normalize' are not specified, affecting the scale of Δt and hence the transition matrix.
axioms (5)
  • domain assumption Linear-Gaussian state-space model: latent user state evolves as x_t = F_t x_{t-1} + noise and semantic observation is a linear projection z_t = M h_static_t
    Section 3, Eq. (10)-(11). This is the core modeling premise of the temporal branch; no empirical comparison to nonlinear dynamics is provided.
  • ad hoc to paper Trace of posterior covariance is a sufficient uncertainty signal for fusion
    Eq. (13) and (16). The paper designs UDFM/ACFM around this scalar; there is no external or theoretical justification beyond the reported ablations.
  • domain assumption Sampled candidate ranking with 1 positive + 100 timestamp-available negatives approximates catalog-scale ranking
    Experiment Setting. All models share this protocol; the paper's SOTA claim is relative to this sampled protocol, not full-corpus ranking.
  • domain assumption Text is the most reliable item-side evidence early in the lifecycle and remains comparable over time
    Introduction and FD-TSE. The framework relies heavily on item text quality; the separate timestamped semantic-evidence audit is promised in the supplementary but not present.
  • domain assumption Titans memory architecture provides the claimed local-attention and long-term memory separation
    Section 2/3, relying on Behrouz et al. 2025. The paper adopts the architecture without re-validating its properties in this setting.
invented entities (1)
  • User-conditioned latent temporal state x_t|t no independent evidence
    purpose: Captures the user's preference state over time, updated by Kalman filtering over history item semantic observations; contributes h_dynamic_t to the ranking representation.
    A latent variable with no direct external measurement; its existence and quality are only evidenced by downstream ranking metrics and internal diagnostics, not by any falsifiable prediction outside the model.

pith-pipeline@v1.3.0-daily-deepseek · 13639 in / 13472 out tokens · 132997 ms · 2026-08-01T15:58:07.115374+00:00 · methodology

0 comments
read the original abstract

Item cold-start recommendation is difficult when new items have sparse early interactions and appear in recommendation environments that keep changing over time. Static content, early feedback, and temporal-state evidence are all useful, but their reliability varies across the item lifecycle. This work proposes a framework--Neural Memory Kalman Fusion Recommender (NMKFR), which combines a Titans-based semantic encoder with time-aware Kalman state tracking. The semantic branch extracts memory-enhanced item observations from text, while the temporal branch estimates latent states under irregular interaction intervals. The NMKFR further uses posterior covariance as an uncertainty signal to calibrate semantic memory retrieval and adaptive static-temporal fusion. Experiments on Amazon Video Games and MovieLens-32M evaluate NMKFR under time-aware and item cold-start protocols using sampled candidate ranking. Across the reported comparisons, ablations, diagnostics, and robustness analyses, NMKFR achieves the strongest retained results and exhibits bounded uncertainty-related internal behavior. These findings provide empirical evidence for posterior-covariance-guided semantic-temporal fusion under the evaluated offline settings.

Figures

Figures reproduced from arXiv: 2607.26429 by Chengzhi Liu, Ning Zeng, Zehui Qu.

Figure 1
Figure 1. Figure 1: Evidence-coordination gap in time-aware item [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the proposed NMKFR framework. FD-TSE encodes item text, TA-KDT performs time-conditioned [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Optimization trajectories of posterior covariance [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 5 canonical work pages

  1. [1]

    Cross-Task Collaborative Meta-Learning for Cold-Start Recommendations , year=

    Du, Yantong and Chen, Rui and Tan, Qiaoyu and Han, Qilong and Wang, Shenjie and Zhao, Xiangyu , journal=. Cross-Task Collaborative Meta-Learning for Cold-Start Recommendations , year=

  2. [3]

    Graph-Structured Driven Dual Adaptation for Mitigating Popularity Bias , year=

    Cai, Miaomiao and Chen, Lei and Wang, Yifan and Cheng, Zhiyong and Zhang, Min and Wang, Meng , journal=. Graph-Structured Driven Dual Adaptation for Mitigating Popularity Bias , year=

  3. [4]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Titans: Learning to Memorize at Test Time , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

  4. [5]

    , title =

    Fan, Ziwei and Liu, Zhiwei and Zhang, Jiawei and Xiong, Yun and Zheng, Lei and Yu, Philip S. , title =. Proceedings of the 30th ACM International Conference on Information & Knowledge Management , pages =. 2021 , isbn =. doi:10.1145/3459637.3482242 , abstract =

    arXiv 2021

  5. [6]

    Proceedings of the ACM Web Conference 2023 , pages =

    Zhou, Zhihui and Zhang, Lilin and Yang, Ning , title =. Proceedings of the ACM Web Conference 2023 , pages =. 2023 , isbn =. doi:10.1145/3543507.3583286 , abstract =

    arXiv 2023

  6. [7]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Preference Aware Dual Contrastive Learning for Item Cold-Start Recommendation , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2024 , month=. doi:10.1609/aaai.v38i8.28763 , abstractNote=

    work page doi:10.1609/aaai.v38i8.28763 2024

  7. [9]

    Proceedings of the 29th ACM International Conference on Multimedia , pages =

    Wei, Yinwei and Wang, Xiang and Li, Qi and Nie, Liqiang and Li, Yan and Li, Xuanping and Chua, Tat-Seng , title =. Proceedings of the 29th ACM International Conference on Multimedia , pages =. 2021 , isbn =. doi:10.1145/3474085.3475665 , abstract =

    arXiv 2021

  8. [11]

    Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Hu, Guoqing and Zhang, An and Liu, Shuo and Cai, Zhibo and Yang, Xun and Wang, Xiang , title =. Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2025 , isbn =. doi:10.1145/3726302.3729894 , abstract =

    arXiv 2025

  9. [12]

    Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Xu, Xiaolong and Dong, Hongsheng and Qi, Lianyong and Zhang, Xuyun and Xiang, Haolong and Xia, Xiaoyu and Xu, Yanwei and Dou, Wanchun , title =. Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2024 , isbn =. doi:10.1145/3626772.3657839 , abstract =

    arXiv 2024

  10. [13]

    2025 , eprint=

    SaviorRec: Semantic-Behavior Alignment for Cold-Start Recommendation , author=. 2025 , eprint=

    2025

  11. [14]

    Proceedings of the Eighteenth ACM International Conference on Web Search and Data Mining , pages =

    Huang, Feiran and Bei, Yuanchen and Yang, Zhenghang and Jiang, Junyi and Chen, Hao and Shen, Qijie and Wang, Senzhang and Karray, Fakhri and Yu, Philip S , title =. Proceedings of the Eighteenth ACM International Conference on Web Search and Data Mining , pages =. 2025 , isbn =. doi:10.1145/3701551.3703546 , abstract =

    arXiv 2025

  12. [15]

    Few-Shot Learning for Cold-Start Recommendation

    Li, Mingming and Hu, Songlin and Zhu, Fuqing and Zhu, Qiannan. Few-Shot Learning for Cold-Start Recommendation. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). 2024

    2024

  13. [16]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Counterfactual Task-augmented Meta-learning for Cold-start Sequential Recommendation , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2025 , month=. doi:10.1609/aaai.v39i12.33396 , abstractNote=

    work page doi:10.1609/aaai.v39i12.33396 2025

  14. [17]

    , title =

    Liu, Huafeng and Zhou, Mingjie and Jing, Liping and Ng, Michael K. , title =. Proceedings of the 31st ACM International Conference on Multimedia , pages =. 2023 , isbn =. doi:10.1145/3581783.3612446 , abstract =

    arXiv 2023

  15. [18]

    Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Huai, Zepeng and Yang, Yuji and Zhang, Mengdi and Zhang, Zhongyi and Li, Yichun and Wu, Wei , title =. Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2023 , isbn =. doi:10.1145/3539618.3591720 , abstract =

    arXiv 2023

  16. [19]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 , pages =

    Zhang, Sheng and Wang, Maolin and Wang, Wanyu and Gao, Jingtong and Zhao, Xiangyu and Yang, Yu and Wei, Xuetao and Liu, Zitao and Xu, Tong , title =. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 , pages =. 2025 , isbn =. doi:10.1145/3690624.3709304 , abstract =

    arXiv 2025

  17. [20]

    2024 , eprint=

    Mamba4Rec: Towards Efficient Sequential Recommendation with Selective State Space Models , author=. 2024 , eprint=

    2024

  18. [21]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 , pages =

    Yoo, Hyunsik and Qiu, Ruizhong and Xu, Charlie and Wang, Fei and Tong, Hanghang , title =. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 , pages =. 2025 , isbn =. doi:10.1145/3690624.3709299 , abstract =

    arXiv 2025

  19. [22]

    Proceedings of the ACM Web Conference 2024 , pages =

    Wang, Lei and Ma, Chen and Wu, Xian and Qiu, Zhaopeng and Zheng, Yefeng and Chen, Xu , title =. Proceedings of the ACM Web Conference 2024 , pages =. 2024 , isbn =. doi:10.1145/3589334.3645400 , abstract =

    arXiv 2024

  20. [23]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 , pages =

    Li, YanHui and Wang, Dongxia and Sun, Zhu and Zhang, Haonan and Guo, Huizhong , title =. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 , pages =. 2025 , isbn =. doi:10.1145/3711896.3737026 , abstract =

    arXiv 2025

  21. [24]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Temporally and Distributionally Robust Optimization for Cold-Start Recommendation , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2024 , month=. doi:10.1609/aaai.v38i8.28721 , abstractNote=

    work page doi:10.1609/aaai.v38i8.28721 2024

  22. [25]

    Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence,

    Preference-Adaptive Meta-Learning for Cold-Start Recommendation , author =. Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence,. 2021 , month =. doi:10.24963/ijcai.2021/222 , url =

    work page doi:10.24963/ijcai.2021/222 2021

  23. [26]

    Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Chen, Hao and Wang, Zefan and Huang, Feiran and Huang, Xiao and Xu, Yue and Lin, Yishi and He, Peng and Li, Zhoujun , title =. Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2022 , isbn =. doi:10.1145/3477495.3531897 , abstract =

    arXiv 2022

  24. [27]

    Evidential Stochastic Differential Equations for Time-Aware Sequential Recommendation , url =

    Neupane, Krishna Prasad and Zheng, Ervine and Yu, Qi , booktitle =. Evidential Stochastic Differential Equations for Time-Aware Sequential Recommendation , url =. 2024 , address =. doi:10.52202/079017-2157 , editor =

    work page doi:10.52202/079017-2157 2024

  25. [28]

    Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Li, Jingyuan and Zhang, Yue and Lin, Xuan and Yang, Xinxing and Zhou, Ge and Li, Longfei and Chen, Hong and Zhou, Jun , title =. Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2023 , isbn =. doi:10.1145/3539618.3592068 , abstract =

    arXiv 2023

  26. [29]

    Firzen: Firing Strict Cold-Start Items with Frozen Heterogeneous and Homogeneous Graphs for Recommendation , year=

    He, Hulingxiao and He, Xiangteng and Peng, Yuxin and Shan, Zifei and Su, Xin , booktitle=. Firzen: Firing Strict Cold-Start Items with Frozen Heterogeneous and Homogeneous Graphs for Recommendation , year=. doi:10.1109/ICDE60146.2024.00354 , publisher=

    arXiv 2024

  27. [30]

    and Chen, Minmin , title =

    Wang, Jianling and Lu, Haokai and Caverlee, James and Chi, Ed H. and Chen, Minmin , title =. Companion Proceedings of the ACM Web Conference 2024 , pages =. 2024 , isbn =. doi:10.1145/3589335.3651532 , abstract =

    arXiv 2024

  28. [31]

    LLMT ree R ec: Unleashing the Power of Large Language Models for Cold-Start Recommendations

    Zhang, Wenlin and Wu, Chuhan and Li, Xiangyang and Wang, Yuhao and Dong, Kuicai and Wang, Yichao and Dai, Xinyi and Zhao, Xiangyu and Guo, Huifeng and Tang, Ruiming. LLMT ree R ec: Unleashing the Power of Large Language Models for Cold-Start Recommendations. Proceedings of the 31st International Conference on Computational Linguistics. 2025

    2025

  29. [32]

    , title =

    Kieu, Hai-Dang and Nguyen, Minh-Duc and Nguyen, Thanh-Son and Le, Dung D. , title =. Companion Proceedings of the ACM on Web Conference 2025 , pages =. 2025 , isbn =. doi:10.1145/3701716.3717855 , abstract =

    arXiv 2025

  30. [33]

    arXiv preprint arXiv:2403.03952 , year=

    Bridging Language and Items for Retrieval and Recommendation , author=. arXiv preprint arXiv:2403.03952 , year=

    Pith/arXiv arXiv

  31. [34]

    Maxwell and Konstan, Joseph A

    Harper, F. Maxwell and Konstan, Joseph A. , title =. ACM Trans. Interact. Intell. Syst. , month = dec, articleno =. 2015 , issue_date =. doi:10.1145/2827872 , abstract =

    doi:10.1145/2827872 2015

  32. [35]

    Proceedings of the 28th ACM International Conference on Information and Knowledge Management , pages =

    Sun, Fei and Liu, Jun and Wu, Jian and Pei, Changhua and Lin, Xiao and Ou, Wenwu and Jiang, Peng , title =. Proceedings of the 28th ACM International Conference on Information and Knowledge Management , pages =. 2019 , isbn =. doi:10.1145/3357384.3357895 , abstract =

    arXiv 2019

  33. [36]

    Self-Attentive Sequential Recommendation , year=

    Kang, Wang-Cheng and McAuley, Julian , booktitle=. Self-Attentive Sequential Recommendation , year=. doi:10.1109/ICDM.2018.00035 , location=

    arXiv 2018

  34. [37]

    arXiv preprint arXiv:2107.10043 , year=

    KalmanNet: Neural Network Aided Kalman Filtering for Partially Known Dynamics , author=. arXiv preprint arXiv:2107.10043 , year=

    Pith/arXiv arXiv

  35. [38]

    arXiv preprint arXiv:2406.15131 , year=

    KalMamba: Towards Efficient Probabilistic State Space Models for RL under Uncertainty , author=. arXiv preprint arXiv:2406.15131 , year=

    Pith/arXiv arXiv

  36. [39]

    Interest-Aware Graph Contrastive Learning for Recommendation With Diffusion-Based Augmentation , year=

    Jing, Mengyuan and Zhu, Yanmin and Wang, Zhaobo and Yu, Jiadi and Tang, Feilong , journal=. Interest-Aware Graph Contrastive Learning for Recommendation With Diffusion-Based Augmentation , year=

  37. [40]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Li, Hourun and Wang, Yifan and Xiao, Zhiping and Yang, Jia and Zhou, Changling and Zhang, Ming and Ju, Wei , title=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=. 2025 , doi=

    2025