DeBaTeR: Denoising Bipartite Temporal Graph for Recommendation

Alyssa Woo; Fei Wang; Hanghang Tong; Jose Sepulveda; Mostafa Rahmani; Xinyu He

REVIEW 3 major objections 6 minor 31 references

DeBaTeR: Denoising Bipartite Temporal Graph for Recommendation

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Adding a learned timestamp vector to user and item embeddings lets graph collaborative filtering detect noisy interactions, improving utility and robustness.

desk verdict Useful item-side temporal denoising result, but the paper's user-side temporal pattern story is algebraically unsupported and needs to be cut or fixed. read the letter →

arxiv 2411.09181 v1 pith:ZT7WQKHX submitted 2024-11-14 cs.IR cs.AIcs.LG

Xinyu He , Jose Sepulveda , Mostafa Rahmani , Alyssa Woo , Fei Wang , Hanghang Tong This is my paper · ORCID

classification cs.IRcs.AIcs.LG

keywords denoisingrecommendersystemsgraphcollaborativefilteringtime-awareembeddingstemporalpatternsimplicitfeedbackbipartiterobustnessnoiseinjection

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

This paper tries to establish that interaction timestamps are a usable signal for cleaning noisy implicit-feedback data in graph collaborative filtering, where clicks or purchases often do not reflect real preference. It proposes DeBaTeR, a mechanism that forms time-aware user and item embeddings by adding the same learned timestamp embedding to each, and then uses the dot product of these time-aware embeddings as a reliability score in two denoising strategies: reweighting the adjacency matrix (DeBaTeR-A) and reweighting the loss function (DeBaTeR-L). The paper's claim is that temporal incoherence, an interaction happening at a time when that item's recorded activity is low, marks noisy edges, and that accounting for this improves accuracy and robustness to injected noise. If true, timestamps become a cheap, already-collected signal for denoising, and the additive embedding trick can be dropped into existing neural graph collaborative filtering models.

What carries the argument

The central object is the additive time-aware embedding pair, $e_u^t = e_u + e_t$ and $e_i^t = e_i + e_t$, where $e_t$ is learned by one-hot encoding each timestamp dimension and concatenating per-dimension embeddings (Eqs. 16-18). The corresponding score $P_{u,i}^t = e_u^\top e_i + e_u^\top e_t + e_i^\top e_t + \|e_t\|^2$ is the time-aware preference used in prediction. The same construction is plugged into a cosine-based reliability score for adjacency-matrix reweighting in DeBaTeR-A and into an MLP weight generator trained by gradient matching in DeBaTeR-L, so the timestamp encoder is the piece that carries the temporal signal through both denoising strategies.

What would settle it

Take a dataset where item popularity is uniform across time but each user has strong, known time-of-day preferences, and compare DeBaTeR against the same backbone without timestamps; if the claimed temporal mechanism is working, ranking accuracy should improve, and if the additive user term is inert, it will not. A second check is to inject noise that follows user-side temporal incoherence (a user's favorite item category moved to the wrong time of day) rather than item-popularity noise; the claimed advantage of DeBaTeR should shrink in that setting.

Watch

Extended reading notes

Core claim

DeBaTeR claims that a timestamp embedding $e_t$, added to both sides of the interaction as $e_u^t = e_u + e_t$ and $e_i^t = e_i + e_t$, turns the prediction score $P_{u,i}^t = (e_u + e_t)^\top (e_i + e_t)$ into a sum of a general preference term and a temporal term. The paper argues that this decomposition lets the model capture temporal patterns of users and items, and it uses the same time-aware embeddings in a cosine reliability score for pruning or downweighting graph edges (DeBaTeR-A) and in an MLP weight generator for reweighting training samples (DeBaTeR-L). On four real-world datasets, one of the two variants reaches the best precision, recall, or NDCG in most settings, beating state-of-the-art graph collaborative filtering and denoising baselines on both vanilla data and data with 20 percent injected noise. The paper presents this as the first use of time information for denoising in neural graph collaborative filtering.

Load-bearing premise

The load-bearing premise is that adding the same learned timestamp vector to both user and item embeddings lets the model exploit temporal patterns on either side, but because $e_u^\top e_t$ is constant across items for a fixed user and timestamp, user-specific patterns like 'horror at night, talk shows in the morning' cannot affect the ranking of items, leaving only item-side temporal popularity to carry the signal.

Editorial extensions

If this is right

If the central claim is correct, recommender systems can treat an interaction's timestamp as evidence about whether the interaction is noisy, without needing any explicit feedback label.
The additive embedding trick is architecture-light: existing graph collaborative filtering models can adopt it by adding the timestamp vector to user and item embeddings before scoring and loss computation.
Temporal coherence becomes a usable pruning signal for the bipartite interaction graph, complementing existing denoising methods that rely on embedding similarity or loss magnitude.
Item-side periodic popularity, such as seasonal items clustering at particular times of year, becomes directly learnable from timestamps and can improve ranking at query time.
The method's robustness gains on data with injected random noise suggest it can partially absorb poisoning attacks that place interactions at arbitrary timestamps.

Reading between the lines

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

A direct reading of the score formula suggests the reported gains are most plausibly carried by item-side temporal popularity: for a fixed user and time, the term $e_u^\top e_t$ is constant across items, so user-specific patterns such as 'this user watches horror at night' cannot change the item ranking.
A natural test of the mechanism is to construct a synthetic dataset with strong user-side temporal preferences but flat item-side popularity; if the additive model is what the paper says it is, it should still show an advantage, and if only item-side terms matter, it will match a timestamp-free baseline.
The discrete per-dimension timestamp encoder could be replaced by continuous time embeddings or by separating user-time and item-time vectors, which would let the model actually express user-specific temporal preferences rather than only item popularity by time.
Temporal reliability is an orthogonal signal to loss-based and agreement-based denoising, so combining DeBaTeR's reliability scores with those families could yield further robustness gains, though the paper does not test such combinations.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

What's actually new here is a modest, useful result: temporal incoherence is a detectable noise signal, and you can inject timestamps into existing denoising frameworks (structure denoising [29] and BOD's loss reweighting [26]) with a simple additive encoder. The RQ3 experiment, which constructs noisy datasets by sampling items against hourly popularity, is the best part of the paper—it tells you what the method actually detects. Don't let the title's 'user-side temporal patterns' fool you: the model's ranking score can't represent them.

The math is straightforward. Eq. (22) gives P^t_{u,i} = (e_u+e_t)^T(e_i+e_t) = e_u^T e_i + e_u^T e_t + e_i^T e_t + ||e_t||^2. For a fixed user u and time t, the e_u^T e_t term is constant across items, so ranking is e_u^T e_i + e_i^T e_t plus a constant. That is static preference plus global item-time popularity. The horror-at-night/talk-show-in-morning example in the introduction requires a user-specific time interaction, which is structurally absent. Section 3.1's claim that Eq. (14) captures both user-side and item-side temporal patterns is therefore an overstatement. The RQ3 experiments only probe item-side patterns, and they support that reading.

Other soft spots are proportionate. The ML-1M ablation is genuinely mixed: DeBaTeR-L with time removed from losses and prediction gets better Precision@10 (0.1667 vs 0.1501) and Recall@10 (0.0477 vs 0.0404), though NDCG@10 is slightly worse. So 'time always helps' is not established. Gains over BOD are often within one standard deviation on several metrics, so the SOTA claim should be softened. The code is promised but not linked in the arXiv version.

What the paper does well: it's clearly written, the setup is fair (multiple datasets, standard baselines, official repos for the denoising baselines), and the timestamp encoder is cheap to add to any dot-product CF model. The RQ3 methodology—varying noise distributions to see which ones are detected—should be copied by other denoising papers.

Bottom line: this is an incremental but real contribution to item-side temporal denoising, and the headline user-side mechanism is unsupported. A synthetic test (flipping a user's item types across time slots) would settle the question. I'd send it to peer review: the item-side finding is worth publishing and the user-side claim is fixable by rewriting the motivation and claims. I'd want a major revision before acceptance.

Referee Report

3 major / 6 minor

Summary. The paper proposes DeBaTeR, a mechanism that adds a learned timestamp embedding to user and item embeddings in neural graph collaborative filtering, and uses the resulting time-aware embeddings in two denoising strategies: reweighting the adjacency matrix (DeBaTeR-A) and reweighting the loss function (DeBaTeR-L). The authors argue that temporal incoherence is a useful signal for identifying noisy interactions, motivated by examples such as a user who watches horror movies at night and talk shows in the morning. Experiments on ML-100K, ML-1M, Yelp, and Amazon compare the proposed methods against general and denoising recommender baselines, including an ablation study and a noise-sampling analysis (RQ3).

Significance. If the proposed mechanism works as claimed, it is a simple, general way to incorporate timestamps into graph collaborative filtering and could improve both recommendation utility and robustness to noisy implicit feedback. The paper's RQ3 experiments provide a falsifiable, external check that item-side temporal coherence is being used for denoising, and the authors state that code is provided. However, the headline user-side temporal-pattern claim is not realized by the model, and the ablation results show that time-aware embeddings do not consistently improve utility. The item-side temporal denoising insight is a useful contribution, but the manuscript needs substantial reframing and additional validation before the broader claims are supported.

major comments (3)

[Section 3.1, Eqs. (14)-(15) and (22)] The claim that the time-aware preference in Eq. (22) captures both user-side and item-side temporal patterns is not supported by the model's functional form. For a fixed user u and timestamp t, the score difference between items i and j is P_{u,i}^t - P_{u,j}^t = (e_u + e_t)^T(e_i - e_j), because e_u^T e_t and ||e_t||^2 are constant across items. Thus the model cannot represent a per-user temporal preference that changes the ranking of items differently for different users; it can only express a global item-time popularity term e_t^T e_i. The introduction's motivating example (a user watches horror at night and talk shows in the morning, so a morning horror viewing is noisy) requires a user-time interaction that is structurally absent from Eq. (15). This is not a tuning issue: the additive architecture cannot realize the stated motivating scenario. The RQ3 experiments in Section 4.4 support item-side temporal coherence, but they do not rescue the user-side claim.
[Section 4.3, Tables 5 and 6] The ablation narrative is contradicted by the ML-1M results for DeBaTeR-L. On the vanilla ML-1M set (Table 5), removing time from losses and prediction improves Precision@10 from 0.1501 to 0.1667 and Recall@10 from 0.0404 to 0.0477; on the noisy ML-1M set (Table 6), Precision@10 improves from 0.1538 to 0.1682 and Recall@10 from 0.0422 to 0.0504. The text in Section 4.3 states that removing time-aware embeddings from loss and prediction leads to an overall worse performance on vanilla datasets, which is not true for this dataset. The paper should either qualify this claim or provide a metric-level explanation (e.g., NDCG is higher with time, while precision/recall are not). As written, the utility benefit of time in DeBaTeR-L is inconsistent and weakens the abstract's assertion that time information 'indeed helps'.
[Tables 2 and 3, Section 4.2] Several of the claimed improvements over the BOD baseline are within one standard deviation, and the paper does not report significance tests. For example, in Table 2 on ML-1M, NDCG@10 is 0.1903 (std 0.0028) for DeBaTeR-A versus 0.1901 (std 0.0025) for BOD, and in Table 3 on ML-1M, NDCG@10 is 0.1865 (std 0.0074) for DeBaTeR-L versus 0.1849 (std 0.0052) for BOD. Similar overlaps occur on Yelp and Amazon for several metrics. The paper's statement that the methods 'outperform state-of-the-art models for both utility and robustness' is therefore not uniformly supported by the reported statistics; the authors should either add significance testing, report effect sizes, or temper the claim to specifically identified metrics and datasets.

minor comments (6)

[Section 4.2, Table 7] Please specify the exact baseline set and averaging procedure for the relative improvement percentages in Table 7, since it is currently unclear whether the percentages are averaged over all datasets, all metrics, or all baselines.
[Section 3.2, Eq. (19)] The reliability score uses zero-layer embeddings e_u^(0) and e_i^(0), while the cited method [29] uses first- and second-layer embeddings; please justify why the unpropagated initial embeddings are sufficient for detecting noisy edges.
[Section 3.3, Eqs. (28)-(29)] Please clarify how the negative-sample weight w_{u,j} is generated; the notation in Eq. (28) suggests W(e_u || e_i || e_t), but Eq. (29) uses w_{u,j} without specifying whether the same function is applied with the negative item embedding in the second argument.
[Algorithm 2, line 6] Algorithm 2 line 6 derives embeddings from Eq. (2), whereas DeBaTeR-A uses the time-aware propagation in Eq. (20); please clarify whether DeBaTeR-L's backbone message passing also uses time-aware embeddings or only the losses and prediction.
[Introduction, Section 1] The statement that this is 'the first work to leverage time information in neural graph collaborative filtering recommender systems' is too broad given the existence of temporal graph recommender methods such as [1]; please restrict the novelty claim to the denoising setting or to the specific mechanism proposed.
[Section 3.2, Eq. (26)] The subscripts in Eq. (26) and (27) are difficult to parse (e.g., u_i_t_ui); please use a consistent notation such as t_{ui} for the timestamp of the positive interaction to distinguish it from generic timestamps.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the time-aware embedding construction is an explicit algebraic ansatz, and the denoising claims are tested against external baselines and a noise-sampling probe; the main weakness is an unsupported representational claim, not a circular reduction.

full rationale

The paper's core construction is not a first-principles derivation but an algebraic ansatz. In Section 3.1, the time-aware score is defined as P^t_{ui} = (e_u+e_t)^T(e_i+e_t) (Eq. 22), and the paper itself shows the decomposition into e_u^T e_i + e_u^T e_t + e_i^T e_t + ||e_t||^2 (Eq. 14). This is an identity by construction, so nothing is 'derived' from a hidden assumption that was not already in the definition. The load-bearing empirical claim, that time information helps identify noisy edges, is evaluated against seven external baselines (Tables 2 and 3) and against four noise-sampling schemes (Table 8, RQ3). The RQ3 experiment is an external behavioral probe: model performance varies with the temporal coherence of the injected noise, which is precisely the kind of evidence that can falsify or support the mechanism rather than reduce it to the definition. The self-citations are building blocks, not circular supports: Section 3.2 says 'We generalize the reliability score function in the previous work [29]', and Section 3.1 says 'we use a simplified variant of [17] to learn timestamp embeddings'. These prior works supply an encoder and a reweighting scheme, but the paper's conclusions do not depend on an unverified theorem from those citations; DeBaTeR-A and DeBaTeR-L are additionally compared with BOD, DeCA, T-CE, and other baselines. The genuine weakness is representational, not circular: for a fixed user u and timestamp t, e_u^T e_t and ||e_t||^2 are constant across items, so item ranking in Eq. (22) carries static user preference plus global item-time popularity e_i^T e_t, and the introduction's user-specific example (horror at night, talk shows in the morning) cannot be expressed in the ranking. That is an overclaim or limitation, and the ML-1M ablations in Tables 5 and 6 are ambiguous, but an unsupported capability claim is not the same as fitting a parameter and renaming it a prediction. No step in the paper reduces by construction to its own inputs.

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

The central empirical claims depend on the timestamp encoder learning useful semantic time patterns, on the additive time-aware score being a sufficient model of temporal preference, on the injected-noise model standing in for real implicit-feedback noise, and on the cosine reliability heuristic. These are assumptions, not derived results.

free parameters (4)

Edge pruning threshold beta (DeBaTeR-A) = 0.35
Chosen by hand; controls which edges are deleted or kept in the reweighted adjacency matrix (Eq. 9).
CL loss weight lambda_1 = 0.2 (DeBaTeR-A), 0.005 (DeBaTeR-L)
Chosen by hand; balances the contrastive loss in Eq. (25).
AU loss weight lambda_2 = 1 for both variants
Chosen by hand; balances the alignment-uniformity loss in Eq. (25).
Contrastive temperature tau = 0.7 (DeBaTeR-A), 0.5 (DeBaTeR-L)
Hyperparameter in the contrastive loss (Eq. 5).

assumptions (4)

domain assumption The learned timestamp embedding e_t will align semantically with item temporal patterns (e.g., Christmas movies near Christmas) so that e_i^T e_t expresses item-time affinity.
Invoked in Section 3.1 right after Eq. (14): 'we expect the embedding of a timestamp t near Christmas to be close to Christmas movies' embeddings.' This is an unproven inductive bias.
ad hoc to paper Additive combination (e_u + e_t) and (e_i + e_t) is a sufficient model of temporal preference.
Eqs. (14)-(15) define the decomposition; the paper asserts capability without deriving it from a preference model. The alternative of a full bilinear temporal model is not explored.
domain assumption Injected synthetic noise (random 20% edges, popularity-proportional sampling) is representative of real implicit-feedback noise.
Section 4.1 constructs noisy datasets this way; conclusions about robustness depend on this noise model.
ad hoc to paper Cosine similarity between time-augmented embeddings is a valid reliability score for identifying noisy edges.
Eq. (19) generalizes the reliability score from [29] by adding e_t; no theoretical justification links cosine similarity to noise probability.

how reviews work

0 comments

Cite this review

Pith. "Pith review of DeBaTeR: Denoising Bipartite Temporal Graph for Recommendation." pith.science (2026). https://pith.science/paper/ZT7WQKHX

@misc{pith2026241109181,
  author       = {Pith},
  title        = {Pith review of: DeBaTeR: Denoising Bipartite Temporal Graph for Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZT7WQKHX}},
  note         = {Machine review of arXiv:2411.09181}
}

read the original abstract

Due to the difficulty of acquiring large-scale explicit user feedback, implicit feedback (e.g., clicks or other interactions) is widely applied as an alternative source of data, where user-item interactions can be modeled as a bipartite graph. Due to the noisy and biased nature of implicit real-world user-item interactions, identifying and rectifying noisy interactions are vital to enhance model performance and robustness. Previous works on purifying user-item interactions in collaborative filtering mainly focus on mining the correlation between user/item embeddings and noisy interactions, neglecting the benefit of temporal patterns in determining noisy interactions. Time information, while enhancing the model utility, also bears its natural advantage in helping to determine noisy edges, e.g., if someone usually watches horror movies at night and talk shows in the morning, a record of watching a horror movie in the morning is more likely to be noisy interaction. Armed with this observation, we introduce a simple yet effective mechanism for generating time-aware user/item embeddings and propose two strategies for denoising bipartite temporal graph in recommender systems (DeBaTeR): the first is through reweighting the adjacency matrix (DeBaTeR-A), where a reliability score is defined to reweight the edges through both soft assignment and hard assignment; the second is through reweighting the loss function (DeBaTeR-L), where weights are generated to reweight user-item samples in the losses. Extensive experiments have been conducted to demonstrate the efficacy of our methods and illustrate how time information indeed helps identifying noisy edges.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 29 canonical work pages

[29]

Toward s robust neural graph collaborative ﬁltering via structure denoising and e mbedding perturba- tion

Haibo Ye, Xinjie Li, Yuan Yao, and Hanghang Tong. Toward s robust neural graph collaborative ﬁltering via structure denoising and e mbedding perturba- tion. 41(3), feb 2023

work page 2023
[26]

Eﬃcient bi-level optimization for recommendation denoisi ng

Zongwei Wang, Min Gao, Wentao Li, Junliang Yu, Linxin Gu o, and Hongzhi Yin. Eﬃcient bi-level optimization for recommendation denoisi ng. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 2502–2511, 2023

work page 2023
[1]

Tempora l graph neural net- works for social recommendation

Ting Bai, Youjie Zhang, Bin Wu, and Jian-Yun Nie. Tempora l graph neural net- works for social recommendation. In 2020 IEEE International Conference on Big Data (Big Data), pages 898–903, 2020

work page 2020
[2]

Continuous-time sequential recommendation with temporal graph collabora- tive transformer

Ziwei Fan, Zhiwei Liu, Jiawei Zhang, Yun Xiong, Lei Zheng , and Philip S Yu. Continuous-time sequential recommendation with temporal graph collabora- tive transformer. In Proceedings of the 30th ACM international conference on information & knowledge management , pages 433–442, 2021

work page 2021
[3]

Poisoning attacks to graph-based recommender systems

Minghong Fang, Guolei Yang, Neil Zhenqiang Gong, and Jia Liu. Poisoning attacks to graph-based recommender systems. In Proceedings of the 34th Annual Computer Security Applications Conference, ACSAC ’18, page 381–392, New York, NY, USA, 2018. Association for Computing Machinery

work page 2018
[4]

Personalized ranking for non-uniformly sampled it ems

Zeno Gantner, Lucas Drumond, Christoph Freudenthaler, and Lars Schmidt- Thieme. Personalized ranking for non-uniformly sampled it ems. In Proceedings of the 2011 International Conference on KDD Cup 2011 - Volume 18, KDDCUP’11, page 231–247. JMLR.org, 2011

work page 2011
[5]

Self-guided learning to denoise for robust recom mendation

Yunjun Gao, Yuntao Du, Yujia Hu, Lu Chen, Xinjun Zhu, Ziqu an Fang, and Bai- hua Zheng. Self-guided learning to denoise for robust recom mendation. In SIGIR, pages 1412–1422, 2022

work page 2022
[6]

Maxwell Harper and Joseph A

F. Maxwell Harper and Joseph A. Konstan. The movielens da tasets: History and context. 5(4), dec 2015

work page 2015

Show all 31 references

[8]

Lightgcn: Simplifying and powering graph convolutio n network for rec- ommendation

Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhan g, and Meng Wang. Lightgcn: Simplifying and powering graph convolutio n network for rec- ommendation. In Proceedings of the 43rd International ACM SIGIR conference o n research and development in Information Retrieval ,...

2020
[9]

Neural collaborative ﬁltering

Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia H u, and Tat-Seng Chua. Neural collaborative ﬁltering. In Proceedings of the 26th international conference on world wide web , pages 173–182, 2017

2017
[10]

Self-attentive se quential recommenda- tion

Wang-Cheng Kang and Julian McAuley. Self-attentive se quential recommenda- tion. In 2018 IEEE international conference on data mining (ICDM), pages 197–206. IEEE, 2018

2018
[11]

Shilling attack models i n recommender system

Parneet Kaur and Shivani Goel. Shilling attack models i n recommender system. In 2016 International Conference on Inventive Computation Te chnologies (ICICT), volume 2, pages 1–5, 2016

2016
[12]

Time2vec: Learning a vector representation of ti me

Seyed Mehran Kazemi, Rishab Goel, Sepehr Eghbali, Jana han Ramanan, Jaspreet Sahota, Sanjay Thakur, Stella Wu, Cathal Smyth, Pascal Poup art, and Marcus Brubaker. Time2vec: Learning a vector representation of ti me. arXiv preprint arXiv:1907.05321, 2019

1907 arXiv
[13]

Improving graph collaborative ﬁltering with neighborhood-enriched contrastive learning

Zihan Lin, Changxin Tian, Yupeng Hou, and Wayne Xin Zhao . Improving graph collaborative ﬁltering with neighborhood-enriched contrastive learning. In Pro- ceedings of the ACM web conference 2022 , pages 2320–2329, 2022

2022
[14]

Rec- ommender system application developments: a survey

Jie Lu, Dianshuang Wu, Mingsong Mao, Wei Wang, and Guang quan Zhang. Rec- ommender system application developments: a survey. Decision support systems, 74:12–32, 2015

2015
[15]

Justifying recommendations using distantly-labeled reviews and ﬁne-grained aspects

Jianmo Ni, Jiacheng Li, and Julian McAuley. Justifying recommendations using distantly-labeled reviews and ﬁne-grained aspects. In Conference on Empirical Methods in Natural Language Processing , 2019

2019
[16]

Eﬀective and eﬃc ient training for se- quential recommendation using recency sampling

Aleksandr Petrov and Craig Macdonald. Eﬀective and eﬃc ient training for se- quential recommendation using recency sampling. In Proceedings of the 16th ACM Conference on Recommender Systems , RecSys ’22, page 81–91, New York, NY, USA, 2022. Association for Computing Machinery

2022
[17]

Incorpo rating time in se- quential recommendation models

Mostafa Rahmani, James Caverlee, and Fei Wang. Incorpo rating time in se- quential recommendation models. In Proceedings of the 17th ACM Conference on Recommender Systems , RecSys ’23, page 784–790, New York, NY, USA, 2023. Association for Computing Machinery

2023
[18]

Bpr: Bayesian personalized ranking from implicit f eedback

Steﬀen Rendle, Christoph Freudenthaler, Zeno Gantner , and Lars Schmidt- Thieme. Bpr: Bayesian personalized ranking from implicit f eedback. In Pro- ceedings of the Twenty-Fifth Conference on Uncertainty in A rtiﬁcial Intelligence, UAI ’09, page 452–461, Arlington, Virginia, U...

2009
[19]

Bert4rec: Sequential recommendation with bidirectional e ncoder representa- tions from transformer

Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu O u, and Peng Jiang. Bert4rec: Sequential recommendation with bidirectional e ncoder representa- tions from transformer. In Proceedings of the 28th ACM international conference on information and knowledge management , p...

2019
[20]

Learning to denoise unreliable interactions for graph coll aborative ﬁltering

Changxin Tian, Yuexiang Xie, Yaliang Li, Nan Yang, and W ayne Xin Zhao. Learning to denoise unreliable interactions for graph coll aborative ﬁltering. SI- GIR ’22, page 122–132, New York, NY, USA, 2022. Association f or Computing Machinery

2022
[22]

Towards representation alignment and uniformity in collaborative ﬁltering

Chenyang Wang, Yuanqing Yu, Weizhi Ma, Min Zhang, Chong Chen, Yiqun Liu, and Shaoping Ma. Towards representation alignment and uniformity in collaborative ﬁltering. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining , pages 1816–1825, 2022

2022
[23]

De- noising implicit feedback for recommendation

Wenjie Wang, Fuli Feng, Xiangnan He, Liqiang Nie, and Ta t-Seng Chua. De- noising implicit feedback for recommendation. In Proceedings of the 14th ACM international conference on web search and data mining , pages 373–381, 2021

2021
[24]

Learning robust recommenders through cross-model agreement

Yu Wang, Xin Xin, Zaiqiao Meng, Joemon M Jose, Fuli Feng, and Xiangnan He. Learning robust recommenders through cross-model agreement. In Proceedings of the ACM Web Conference 2022 , pages 2015–2025, 2022

2022
[25]

Implicit feedbacks are not always favorable: Iterative rel abeled one-class col- laborative ﬁltering against noisy interactions

Zitai Wang, Qianqian Xu, Zhiyong Yang, Xiaochun Cao, an d Qingming Huang. Implicit feedbacks are not always favorable: Iterative rel abeled one-class col- laborative ﬁltering against noisy interactions. In Proceedings of the 29th ACM International Conference on Multimedia, MM ...

2021
[27]

Self-supervised graph learning for recommend ation

Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Ch en, Jianxun Lian, and Xing Xie. Self-supervised graph learning for recommend ation. In Proceed- ings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’21, page 72...

2021
[28]

Contrastive learning for sequential recommendation

Xu Xie, Fei Sun, Zhaoyang Liu, Shiwen Wu, Jinyang Gao, Ji andong Zhang, Bolin Ding, and Bin Cui. Contrastive learning for sequential recommendation. In 2022 IEEE 38th international conference on data engineering (IC DE), pages 1259–1273. IEEE, 2022

2022
[30]

Are graph augmentations necessary? simple graph co ntrastive learn- ing for recommendation

Junliang Yu, Hongzhi Yin, Xin Xia, Tong Chen, Lizhen Cui , and Quoc Viet Hung Nguyen. Are graph augmentations necessary? simple graph co ntrastive learn- ing for recommendation. In Proceedings of the 45th international ACM SIGIR conference on research and development in inform...

2022
[31]

Self- supervised learning for recommender systems: A survey

Junliang Yu, Hongzhi Yin, Xin Xia, Tong Chen, Jundong Li , and Zi Huang. Self- supervised learning for recommender systems: A survey. IEEE Transactions on Knowledge and Data Engineering , 2023

2023
[32]

Data poisoning attack against recommender sys tem using incom- plete and perturbed data

Hengtong Zhang, Changxin Tian, Yaliang Li, Lu Su, Nan Ya ng, Wayne Xin Zhao, and Jing Gao. Data poisoning attack against recommender sys tem using incom- plete and perturbed data. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , KDD ’21, ...

2021
[33]

Robust recommender system: A survey and future directions

Kaike Zhang, Qi Cao, Fei Sun, Yunfan Wu, Shuchang Tao, Hu awei Shen, and Xueqi Cheng. Robust recommender system: A survey and future directions. arXiv preprint arXiv:2309.02057 , 2023

2023 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [29]

Toward s robust neural graph collaborative ﬁltering via structure denoising and e mbedding perturba- tion

Haibo Ye, Xinjie Li, Yuan Yao, and Hanghang Tong. Toward s robust neural graph collaborative ﬁltering via structure denoising and e mbedding perturba- tion. 41(3), feb 2023

work page 2023

[2] [26]

Eﬃcient bi-level optimization for recommendation denoisi ng

Zongwei Wang, Min Gao, Wentao Li, Junliang Yu, Linxin Gu o, and Hongzhi Yin. Eﬃcient bi-level optimization for recommendation denoisi ng. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 2502–2511, 2023

work page 2023

[3] [1]

Tempora l graph neural net- works for social recommendation

Ting Bai, Youjie Zhang, Bin Wu, and Jian-Yun Nie. Tempora l graph neural net- works for social recommendation. In 2020 IEEE International Conference on Big Data (Big Data), pages 898–903, 2020

work page 2020

[4] [2]

Continuous-time sequential recommendation with temporal graph collabora- tive transformer

Ziwei Fan, Zhiwei Liu, Jiawei Zhang, Yun Xiong, Lei Zheng , and Philip S Yu. Continuous-time sequential recommendation with temporal graph collabora- tive transformer. In Proceedings of the 30th ACM international conference on information & knowledge management , pages 433–442, 2021

work page 2021

[5] [3]

Poisoning attacks to graph-based recommender systems

Minghong Fang, Guolei Yang, Neil Zhenqiang Gong, and Jia Liu. Poisoning attacks to graph-based recommender systems. In Proceedings of the 34th Annual Computer Security Applications Conference, ACSAC ’18, page 381–392, New York, NY, USA, 2018. Association for Computing Machinery

work page 2018

[6] [4]

Personalized ranking for non-uniformly sampled it ems

Zeno Gantner, Lucas Drumond, Christoph Freudenthaler, and Lars Schmidt- Thieme. Personalized ranking for non-uniformly sampled it ems. In Proceedings of the 2011 International Conference on KDD Cup 2011 - Volume 18, KDDCUP’11, page 231–247. JMLR.org, 2011

work page 2011

[7] [5]

Self-guided learning to denoise for robust recom mendation

Yunjun Gao, Yuntao Du, Yujia Hu, Lu Chen, Xinjun Zhu, Ziqu an Fang, and Bai- hua Zheng. Self-guided learning to denoise for robust recom mendation. In SIGIR, pages 1412–1422, 2022

work page 2022

[8] [6]

Maxwell Harper and Joseph A

F. Maxwell Harper and Joseph A. Konstan. The movielens da tasets: History and context. 5(4), dec 2015

work page 2015