TSRec: Enhancing Repeat-Aware Recommendation from a Temporal-Sequential Perspective

Fan Wu; Shigang Quan; Shui Liu; Zhenzhe Zheng

REVIEW 2 major objections 8 minor 40 references

TSRec: Enhancing Repeat-Aware Recommendation from a Temporal-Sequential Perspective

T0 review · 2 major / 8 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read TSRec claims that modeling repeat-consumption timing and sequence context yields state-of-the-art next-item prediction, reporting gains up to 5.47% over ten baselines on three benchmarks.

desk verdict Useful architecture with a likely temporal leak in the item-side matrix; the headline gains are not believable until the authors confirm M_i is built from training data only. read the letter →

arxiv 2506.08531 v1 pith:6RVJMRPF submitted 2025-06-10 cs.IR

Shigang Quan , Shui Liu , Zhenzhe Zheng , Fan Wu This is my paper

classification cs.IR

keywords repeat-awarerecommendationtemporalpatternmodelingsequencematchingrepeatconsumptiontimeintervalmatrixnext-itempredictionsequentialrecommendersystems

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 repeat consumption is driven by two learnable regularities: users tend to re-engage with an item after characteristic time intervals, and a repeat event is typically preceded by a behavior sequence similar to the one before an earlier repeat. TSRec encodes these regularities in three modules: a user-specific temporal attention module over the user's own repeat intervals, an item-specific module that pools repeat intervals from other users into a Repeat Time Interval Matrix, and a sequence-matching module that aligns the current behavior sequence with the last repeat sequence. The authors report that TSRec outperforms ten baselines, both general sequential recommenders and prior repeat-aware methods, on RetailRocket, LastFM, and Diginetica across HR@5/10/20 and NDCG@5/10/20, with the largest gain being 5.47% in HR@5 on Diginetica. If the claim holds, explicitly modeling repeat intentions improves next-item prediction without sacrificing new-item accuracy.

What carries the argument

The load-bearing object is the Repeat Time Interval Matrix $M_i\in\mathbb{N}^{m\times n}$: for a target item $i$, each row is one user's sequence of discretized repeat intervals $p^{u,i}_k=\lfloor |t^{u,i}_{k+1}-t^{u,i}_k|/p_{\min}\rfloor$, so the matrix aggregates repeat-timing knowledge across many users. TSRec embeds the matrix and runs multi-scale 2D convolutions with mean and max pooling in ITRM, letting the model borrow temporal regularity from other users when the target user's own repeat history is sparse. UTRM carries the user-specific signal by applying scaled dot-product attention with the target interval as query and the user's historical intervals as keys and values, while SRAM encodes the current and last-repeat behavior sequences with Bi-GRU and 1D convolution and aligns them through soft attention, element-wise subtraction and product, and gated fusion. Together these modules produce the user representation that is scored against the target item.

What would settle it

Train TSRec with the rows of the Repeat Time Interval Matrix shuffled (or replaced by intervals of a different item) and measure HR@5 on the three datasets; if performance does not degrade materially, the cross-user temporal matrix is not the source of the reported gains, whereas a sharp drop would confirm the matrix is doing real work.

Watch

Extended reading notes

Core claim

The central claim is that the temporal and sequential patterns behind repeat consumption are complementary, and a recommender that captures both can outperform models that treat repeat and new items uniformly. The paper demonstrates this with TSRec, whose UTRM uses scaled dot-product attention to relate the target time interval to the user's historical repeat intervals, whose ITRM extracts item-level temporal regularity from a Repeat Time Interval Matrix built from many users' repeat intervals, and whose SRAM encodes the current and last-repeat behavior sequences with Bi-GRU and convolution, then aligns them with soft attention, element-wise differences and products, and a gated fusion. Across RetailRocket, LastFM, and Diginetica, the paper reports consistent state-of-the-art results on all twelve HR/NDCG metrics at K=5,10,20, with improvements over the best baseline ranging from 0.62% to 5.47% depending on metric and dataset. The intended upshot is that repeat-aware recommendation should be modeled as a joint temporal-sequential problem rather than as a single probability of repetition.

Load-bearing premise

The Repeat Time Interval Matrix assumes that an item's repeat time intervals are distributed similarly across users, so intervals from other users can stand in for the target user's sparse repeat history; if repeat cadence varies strongly across users, this side information becomes noise and the item-temporal module's reported gains would not generalize.

Editorial extensions

If this is right

The reported results imply that repeat-aware models can beat general sequential recommenders even on datasets with low repeat ratios, since RetailRocket (35.03% repeats) and Diginetica (16.53% repeats) still show TSRec ahead of SASRec and BERT4Rec.
Ablations show all three modules contribute, with removing ITRM costing the most, so cross-user repeat-interval side information is the largest single source of gain under the paper's setup.
The split evaluation on new versus repeat items indicates that temporal-sequential repeat modeling does not simply over-fit to repeat items; TSRec reports higher accuracy than baselines on both categories on RetailRocket.
The case studies suggest the model assigns higher scores to intervals near the user's actual last repeat interval, which would make it useful for predicting when, not just whether, a repeat occurs.

Reading between the lines

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

Inference: the paper's assumption of uniform repeat intervals across users could be tested and exploited by clustering users by interval distribution; if clusters differ, per-cluster matrices would likely improve on the single global $M_i$, a variant the paper does not explore.
Inference: the Repeat Time Interval Matrix format suggests a cold-start transfer: an item's matrix learned on one user population could seed recommendations for new users who have interacted with that item once, though the paper does not report any cross-user or cold-start experiment.
Inference: because the Jaccard overlap between current and last-repeat sequences is 45-60%, a simpler retrieval baseline that recommends items from the most similar past repeat segment might capture part of the SRAM gains; the paper does not compare against such a baseline.
Inference: a natural stress test is to apply TSRec to a domain with very infrequent repeat events, where the temporal modules' contribution may shrink; reporting per-dataset contributions with repeat-ratio controls would clarify when the model's advantage is largest.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The thing to know about arXiv:2506.08531 is that the model is a coherent assembly of known pieces, and the evaluation is broad, but the headline numbers may be inflated by a temporal leak in the item-side module. The authors never state that the Repeat Time Interval Matrix M_i is built from the training split only. As written, Definition 3.1 and §4.1.4 allow M_i to include every repeat interval for an item across all users, with no cutoff at the prediction timestamp. If that is what the code does, then for any target user the matrix can contain the target user's own future repeat intervals for the item being scored, and the ITRM's 2D convolutions can read the label. The ablation shows ITRM removal hurts more than the other modules, so this is not a peripheral detail.

What is genuinely new: TSRec combines target attention over the user's own repeat intervals, a CNN over a cross-user repeat-interval matrix, and a Bi-GRU plus soft-alignment matching between the current behavior sequence and the last repeat sequence. Each technique is known, but the specific composition for repeat-aware recommendation is new and the ablation demonstrates each component contributes. Ten baselines, three public datasets, and HR/NDCG at 5/10/20 is solid empirical scope.

The soft spots are concentrated in the write-up. Eq. (28) has a misprinted negative-sample term ('-(1 - log(y))' is not a loss); Figure 3 references an undefined 'SPRec' model; and the significance claims come from a paired t-test without reporting variance or number of seeds. The motivation analysis in Section 1 is run on the full datasets, so it is not independent evidence for the uniformity assumption behind M_i. The related-work section also skips the recent time-interval-aware sequential models (TiSASRec and others), which makes the novelty claim appear stronger than it is.

The first thing a referee should demand is a clear statement: M_i is constructed from training-set intervals only, and for each target time the target user's row is truncated to intervals before that time. If the authors can confirm that, the paper is a reasonable contribution. If they cannot, the reported SOTA improvements are a validation of leakage, not of temporal-sequential repeat modeling.

Bottom line: this deserves peer review rather than desk rejection, because the architecture and ablations are worth engaging with and the flaw, if real, is fixable. I would not cite the numbers until the M_i question is answered.

Referee Report

2 major / 8 minor

Summary. The paper proposes TSRec, a repeat-aware recommendation model that combines three modules: a User-specific Temporal Representation Module (UTRM) using target attention over a user's historical repeat time intervals, an Item-specific Temporal Representation Module (ITRM) applying 2D convolutions to a Repeat Time Interval Matrix constructed from other users' repeat intervals, and a Sequential Repeat-Aware Module (SRAM) that aligns the user's current behavior sequence with the last repeat sequence via Bi-GRU, convolution, and attention-style matching. The model is evaluated on RetailRocket, LastFM, and Diginetica against ten baselines, reporting consistent improvements on HR@5/10/20 and NDCG@5/10/20, with ablations showing each module contributes. The central claim is that TSRec achieves state-of-the-art repeat-aware recommendation performance by exploiting temporal and sequential patterns.

Significance. If the central claim is valid, TSRec provides a competitive and well-motivated architecture for repeat-aware recommendation, with simple intuitions (repeat intervals and sequence-level repetition) translated into concrete modules. The paper includes ablations, hyper-parameter studies, and case studies, and the anonymous code link is a useful reproducibility asset. However, the soundness of the empirical claim hinges on a critical detail that is currently unspecified: whether the Repeat Time Interval Matrix used by the ITRM is constructed only from training data or whether it can encode future/test information. Because the reported gains are modest in several metrics and the ITRM ablation shows a clear drop when removed, a temporal-leakage issue would invalidate the headline result. The paper is otherwise readable and the module-level breakdown is clear, but this load-bearing gap must be resolved before the SOTA claim can be accepted.

major comments (2)

[§3.3, Definition 3.1, §4.1.4] The construction of the Repeat Time Interval Matrix M_i is not specified to be restricted to the training split or to intervals occurring before the prediction timestamp. Definition 3.1 stacks P_{u,i} over users without any temporal qualification, and §4.1.4 states only that users are sorted by item repeat frequencies and the Top-M (M=10) are selected. If M_i contains the target user's own future repeat intervals or other users' intervals from the test period, then the ITRM representation used in Eq. (8) can encode label information, and the consistent gains in Table 3 would be an artifact of leakage rather than evidence for temporal modeling. Please state explicitly that M_i is built only from training interactions with an appropriate timestamp cutoff, and if this is already the case, describe the construction precisely enough to be reproduced; otherwise the reported SOTA claim is not supported.
[§4.1.4 and Table 3] The evaluation protocol uses 3 sampled negatives in validation and 100 sampled negatives in testing, but the manuscript does not state whether all ten baselines were trained and evaluated under exactly the same negative-sampling procedure, data splits, early-stopping criterion, and hyper-parameter search. Since the reported improvements are as small as 0.62% (NDCG@20 on RetailRocket), small protocol differences could change the relative ranking. Please provide the shared evaluation setup (or per-baseline setup) and report standard deviations or confidence intervals for the main results so that the 'consistently outperforms' claim can be verified.

minor comments (8)

[Eq. (28)] The negative-sampling term is written as $-(1 - \log \hat{y}(u,v^-,t))$, which is not the standard cross-entropy term (the usual form is $-\log(1 - \hat{y})$) and appears to be a misprint; please correct the objective or state the exact loss used in the implementation.
[Figure 3 and §4.2.2] The legend of Figure 3 includes 'SPRec', which is not defined in §4.1.3 and is not mentioned in the text; please replace it with the intended baseline or define it explicitly.
[§1] The sentence 'Therefore we propose UTRM, which incorporates item time interval information from other users as side information' appears to refer to ITRM, not UTRM; please fix the module name.
[§3.4.1] After Eq. (14), the text says 'we can obtain the Bi-GRU encoded representation matrix for E_l, named H_b'; H_b should be the encoding of E_b, not E_l.
[Definition 3.1 and §4.1.4] Definition 3.1 treats all P_{u,i} as having the same length n, but in practice different users have different numbers of repeat intervals for an item; please specify how M_i is padded, truncated, or otherwise made rectangular.
[Table 3] No standard deviations or confidence intervals are reported despite the claim of statistical significance at p < 0.01; please provide the number of runs and variance estimates for the main results.
[§3.3 and Figure 1] The uniformity claim that 'the item repeat time intervals for different users were relatively uniform' is supported only by Figure 1, which shows one item and one user-item pair; please add aggregate evidence across items and users or soften the claim.
[§4.4.1] There is a stray quotation mark after 'we designate L = 10 as the hyperparameter for TSRec.'; please remove it.

Circularity Check

1 steps flagged · score 6.0 of 10

Temporal leakage in the Repeat Time Interval Matrix makes the SOTA claim partially circular: as written, M_i is built from all users' repeat intervals with no train/test cutoff, so the repeat intervals being predicted can be fed back into the model input.

self definitional [Section 3.3, Definition 3.1 / Eq. (2); construction in Section 4.1.4; score Eq. (27)]
"The Repeat Time Interval Matrix of item i can be defined as M_i∈N^{m×n}: [P_{u1,i}; P_{u2,i}; ...; P_{um,i}] ... To construct the Repeat Time Interval Matrix, we sort each user by item repeat frequencies and choose the Top-M (M=10) users. ... 𝒚(𝑢,𝑣,𝑡)=𝜎(𝒐^T· 𝒆𝑣))."

Eq. (2) defines M_i from P_{u,i}, the repeat-interval sequences of users for item i, without restricting the timestamps to the training split or to intervals ending before the prediction time. For a test repeat of item i by a top-M user u at time t, p_{u,i}=t-last_time is an element of P_{u,i}, so the label's repeat interval is contained in M_i. ITRM embeds M_i as E_P and Eq. (6)-(8) pool it into o_IT; Eq. (26)-(27) concatenate o_IT into o and dot it with the target item to produce the score. Thus the reported SOTA HR/NDCG improvements for repeat items can be obtained in part from the very test repeat intervals being predicted.

full rationale

Most of TSRec is a standard neural assembly (target attention, Bi-GRU+Conv encoders, soft alignment, gated fusion) and the paper cites prior work for these building blocks; there are no load-bearing self-citations and no fitted parameter is relabeled as a prediction. The benchmark comparisons use held-out test splits and are externally grounded. The one structural problem is the Repeat Time Interval Matrix: as defined in Eq. (2), it aggregates P_{u,i} for all users with no timestamp cutoff, and the construction paragraph in Section 4.1.4 does not say it is limited to the training split. Since ITRM feeds this matrix into the final score (Eqs. (6)-(8), (26)-(27)), a test repeat's own time interval can be an input to its prediction. That makes the central SOTA claim partially circular by construction, although the extent would depend on how often the target user is among the top-M users per item and on whether the authors intended an implicit training-only construction. If the matrix were restricted to past interactions, this step would vanish and the score would be 0-2.

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

The central claim rests on several hand-picked hyperparameters (M, L, d, kernel sizes) and two unproven domain assumptions about cross-user temporal uniformity and the transferability of NLP sequence matching. The discretization constant p_min is never reported, which is a reproducibility gap.

free parameters (5)

M (number of users in Repeat Time Interval Matrix) = 10
Chosen by hand as 'the Top-M (M=10) users' in Section 4.1.4; the central ITRM module depends on this choice.
Sequence length L = 10
Grid-searched from {5,10,15,20} on validation in Section 4.4.1; all modules use truncated sequences of this length.
Embedding dimension d = 100
Grid-searched from {10,20,50,100,200} in Section 4.4.2.
Conv2D kernel sizes and count = C=3; kernels 1x1, 1x3, 3x1
Set in Section 4.1.4; defines the scales of temporal patterns extracted by ITRM.
p_min (time interval discretization unit) = not reported
Definition 3.1 discretizes intervals as floor(|difference| / p_min); p_min is never specified in the paper, yet all temporal features are quantized by it.

assumptions (4)

domain assumption Repeat time intervals for an item are relatively uniform across users, so cross-user intervals are informative for a given user.
Invoked in Section 3.3 to justify ITRM; supported only by Figure 1, which shows one item and one user-item pair, not a systematic analysis.
domain assumption The sequence matching and alignment framework from NLP sentence matching transfers to user behavior sequences.
Section 3.4 applies the matching-aggregation framework (citations [6,7]) to behavior sequences without validating that the assumptions of sentence alignment, such as word-level compositionality, hold for interaction sequences.
domain assumption Time-based split (70/10/20) creates i.i.d. evaluation conditions for next-item prediction.
Section 4.1.1 splits by time; the paper does not analyze temporal drift or popularity shifts between splits, which can inflate or deflate repeat-aware model gains.
domain assumption The negative sampling protocol with 3 and 100 negatives yields unbiased comparison across baselines.
Section 4.1.4 states the protocol; no analysis of variance or sensitivity to the number of negatives is reported.

how reviews work

0 comments

Cite this review

Pith. "Pith review of TSRec: Enhancing Repeat-Aware Recommendation from a Temporal-Sequential Perspective." pith.science (2026). https://pith.science/paper/6RVJMRPF

@misc{pith2026250608531,
  author       = {Pith},
  title        = {Pith review of: TSRec: Enhancing Repeat-Aware Recommendation from a Temporal-Sequential Perspective},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6RVJMRPF}},
  note         = {Machine review of arXiv:2506.08531}
}

read the original abstract

Repeat consumption, such as repurchasing items and relistening songs, is a common scenario in daily life. To model repeat consumption, the repeat-aware recommendation has been proposed to predict which item will be re-interacted based on the user-item interactions. In this paper, we investigate various inherent characteristics to enhance the repeat-aware recommendation. Specifically, we explore these characteristics from two aspects: one is from the temporal aspect where we consider the time interval relationship in the user behavior sequence; the other is from the sequential aspect where we consider the sequential-level relationship in the user behavior sequence. And our intuition is that both the temporal pattern and sequential pattern will reflect users' intentions of repeat consumption. By utilizing these two patterns, a novel model called Temporal and Sequential repeat-aware Recommendation(TSRec for short) is proposed to enhance repeat-aware recommendation. TSRec has three main components: 1) User-specific Temporal Representation Module (UTRM), which encodes and extracts user historical repeat temporal information. 2)Item-specific Temporal Representation Module (ITRM), which incorporates item time interval information as side information to alleviate the data sparsity problem of user repeat behavior sequence. 3) Sequential Repeat-Aware Module (SRAM), which represents the similarity between the user's current and the last repeat sequences. Extensive experimental results on three public benchmarks demonstrate the superiority of TSRec over state-of-the-art methods. The implementation code is available https://anonymous.4open.science/r/TSRec-2306/.

Figures

Figures reproduced from arXiv: 2506.08531 by the authors.

**Figure 1.** Data analyses to illustrate the inherent characteristics related to repeat consumption from the temporal aspect on [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** The overview architecture of TSRec, which consists of three parts: User-specific Temporal Representation Module [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** The performance of TSRec is evaluated on both new [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: The impart of hyper-parameters, namely 𝐿 and 𝑑 on two datasets in terms of HR@20 and NDCG@20. 4.5 Case Study of Temporal Modeling(RQ4) In order to verify that TSRec has learned the temporal information of user repeat consumption, we randomly selected a few test cases. …

**Figure 5.** Figure 5: Test cases to verify that TSRec has indeed learned [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 36 canonical work pages

[1]

Ashton Anderson, Ravi Kumar, Andrew Tomkins, and Sergei Vassilvitskii. 2014. The dynamics of repeat consumption. In 23rd International World Wide Web Conference, WWW ’14, Seoul, Republic of Korea, April 7-11, 2014, Chin-Wan Chung, Andrei Z. Broder, Kyuseok Shim, and Torsten Suel (Eds.). ACM, 419–430

work page 2014
[2]

Mozhdeh Ariannezhad, Sami Jullien, Ming Li, Min Fang, Sebastian Schelter, and Maarten de Rijke. 2022. ReCANet: A Repeat Consumption-Aware Neural Network for Next Basket Recommendation in Grocery Shopping. In SIGIR ’22: The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, Madrid, Spain, July 11 - 15, 2022 . 1240–1250

work page 2022
[3]

Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2015. Neural Machine Translation by Jointly Learning to Align and Translate. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings , Yoshua Bengio and Yann LeCun (Eds.)

work page 2015
[4]

Benson, Ravi Kumar, and Andrew Tomkins

Austin R. Benson, Ravi Kumar, and Andrew Tomkins. 2016. Modeling User Consumption Sequences. In Proceedings of the 25th International Conference on World Wide Web, WWW 2016, Montreal, Canada, April 11 - 15, 2016 , Jacqueline Bourdeau, Jim Hendler, Roger Nkambou, Ian Horrocks, and Ben Y. Zhao (Eds.). ACM, 519–529

work page 2016
[5]

Rahul Bhagat, Srevatsan Muralidharan, Alex Lobzhanidze, and Shankar Vish- wanath. 2018. Buy It Again: Modeling Repeat Purchase Recommendations. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Dis- covery & Data Mining, KDD 2018, London, UK, August 19-23, 2018 , Yike Guo and Faisal Farooq (Eds.). ACM, 62–70

work page 2018
[6]

Qian Chen, Zhen-Hua Ling, and Xiaodan Zhu. 2018. Enhancing Sentence Embed- ding with Generalized Pooling. In Proceedings of the 27th International Conference on Computational Linguistics, COLING 2018, Santa Fe, New Mexico, USA, Au- gust 20-26, 2018 , Emily M. Bender, Leon Derczynski, and Pierre Isabelle (Eds.). Association for Computational Linguistics, 1815–1826

work page 2018
[7]

Qian Chen, Xiaodan Zhu, Zhen-Hua Ling, Si Wei, Hui Jiang, and Diana Inkpen

work page
[8]

Xu Chen, Hongteng Xu, Yongfeng Zhang, Jiaxi Tang, Yixin Cao, Zheng Qin, and Hongyuan Zha. 2018. Sequential Recommendation with User Memory Networks. In Proceedings of the Eleventh ACM International Conference on Web Search and Data Mining, WSDM 2018, Marina Del Rey, CA, USA, February 5-9, 2018 , Yi Chang, Chengxiang Zhai, Yan Liu, and Yoelle Maarek (Eds.)...

work page 2018

Show all 40 references

[9]

Kyunghyun Cho, Bart van Merrienboer, Çaglar Gülçehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation. In Proceedings of the 2014 Conference on Empirical Me...

2014
[10]

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Associa- tion for Computational Linguistics: Hum...

2019
[11]

ALAN G. HAWKES. 1971. Spectra of some self-exciting and mutually exciting point processes. Biometrika 58, 1 (1971), 83–90

1971
[12]

Ruining He and Julian J. McAuley. 2016. Fusing Similarity Models with Markov Chains for Sparse Sequential Recommendation. In IEEE 16th International Confer- ence on Data Mining, ICDM 2016, December 12-15, 2016, Barcelona, Spain, Francesco Bonchi, Josep Domingo-Ferrer, Ricardo ...

2016
[13]

Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk
[14]

Balázs Hidasi, Massimo Quadrana, Alexandros Karatzoglou, and Domonkos Tikk. 2016. Parallel Recurrent Neural Network Architectures for Feature-rich Session-based Recommendations. In Proceedings of the 10th ACM Conference on Recommender Systems, Boston, MA, USA, September 15-19,...

2016
[15]

Wang-Cheng Kang and Julian J. McAuley. 2016. Self-Attentive Sequential Rec- ommendation. In IEEE International Conference on Data Mining, ICDM 2018, Singapore, November 17-20, 2018 . 197–206

2016
[16]

Schrater

Komal Kapoor, Karthik Subbian, Jaideep Srivastava, and Paul R. Schrater. 2015. Just in Time Recommendations: Modeling the Dynamics of Boredom in Activity Streams. In Proceedings of the Eighth ACM International Conference on Web Search and Data Mining, WSDM 2015, Shanghai, Chin...

2015
[17]

Yoon Kim. 2014. Convolutional Neural Networks for Sentence Classification. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, EMNLP 2014, October 25-29, 2014, Doha, Qatar, A meeting of SIGDAT, a Special Interest Group of the ACL , Alessa...

2014
[18]

Yehuda Koren and Robert M. Bell. 2015. Advances in Collaborative Filtering. In Recommender Systems Handbook , Francesco Ricci, Lior Rokach, and Bracha Shapira (Eds.). Springer, 77–118

2015
[19]

Jing Li, Pengjie Ren, Zhumin Chen, Zhaochun Ren, Tao Lian, and Jun Ma. 2017. Neural Attentive Session-based Recommendation. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, CIKM 2017, Singapore, November 06 - 10, 2017 . 1419–1428

2017
[20]

Yuncong Li, Cunxiang Yin, Yancheng He, Guoqiang Xu, Jing Cai, Leeven Luo, and Sheng-hua Zhong. 2022. Modeling User Repeat Consumption Behavior for Online Novel Recommendation. In RecSys ’22: Sixteenth ACM Conference on Recommender Systems, Seattle, W A, USA, September 18 - 23,...

2022
[21]

Min Lin, Qiang Chen, and Shuicheng Yan. 2014. Network In Network. In 2nd International Conference on Learning Representations, ICLR 2014, Banff, AB, Canada, April 14-16, 2014, Conference Track Proceedings , Yoshua Bengio and Yann LeCun (Eds.)

2014
[22]

Yang Liu, Matt Gardner, and Mirella Lapata. 2018. Structured Alignment Net- works for Matching Sentences. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, October 31 - November 4, 2018, Ellen Riloff, David Chiang, Ju...

2018
[23]

McAuley, and Karl Aberer

Jérémie Rappaz, Julian J. McAuley, and Karl Aberer. 2021. Recommendation on Live-Streaming Platforms: Dynamic Availability and Repeat Consumption. In RecSys ’21: Fifteenth ACM Conference on Recommender Systems, Amsterdam, The Netherlands, 27 September 2021 - 1 October 2021 , H...

2021
[24]

Markus Reiter-Haas, Emilia Parada-Cabaleiro, Markus Schedl, Elham Motamedi, Marko Tkalcic, and Elisabeth Lex. 2021. Predicting Music Relistening Behavior Using the ACT-R Framework. In RecSys ’21: Fifteenth ACM Conference on Rec- ommender Systems, Amsterdam, The Netherlands, 27...

2021
[25]

Pengjie Ren, Zhumin Chen, Jing Li, Zhaochun Ren, Jun Ma, and Maarten de Rijke. 2019. RepeatNet: A Repeat Aware Neural Recommendation Machine for Session-Based Recommendation. In The Thirty-Third AAAI Conference on Artifi- cial Intelligence, AAAI 2019, The Thirty-First Innovati...

2019
[26]

Steffen Rendle, Christoph Freudenthaler, and Lars Schmidt-Thieme. 2010. Factor- izing personalized Markov chains for next-basket recommendation. InProceedings of the 19th International Conference on World Wide Web, WWW 2010, Raleigh, North Carolina, USA, April 26-30, 2010 , Mi...

2010
[27]

Rumelhart, Geoffrey E

David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams. 1988. Learning Representations by Back-Propagating Errors . MIT Press, Cambridge, MA, USA, 696–699

1988
[28]

Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang
[29]

Yong Kiam Tan, Xinxing Xu, and Yong Liu. 2016. Improved Recurrent Neural Networks for Session-based Recommendations. InProceedings of the 1st Workshop on Deep Learning for Recommender Systems, DLRS@RecSys 2016, Boston, MA, USA, September 15, 2016 , Alexandros Karatzoglou, Balá...

2016
[30]

Gomez, Lukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 20...

2017
[31]

Chenyang Wang, Min Zhang, Weizhi Ma, Yiqun Liu, and Shaoping Ma. 2019. Modeling Item-Specific Temporal Dynamics of Repeat Consumption for Recom- mender Systems. In The World Wide Web Conference, WWW 2019, San Francisco, CA, USA, May 13-17, 2019 . 1977–1987

2019
[32]

Wei Wang, Chen Wu, and Ming Yan. 2018. Multi-Granularity Hierarchical Attention Fusion Networks for Reading Comprehension and Question Answering. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, ACL 2018, Melbourne, Australia, July 15...

2018
[33]

Hongzhi Yin, Bin Cui, Ling Chen, Zhiting Hu, and Xiaofang Zhou. 2015. Dynamic User Modeling in Social Media Systems. ACM Trans. Inf. Syst. 33, 3 (2015), 10:1– 10:44

2015
[34]

Jose, and Xiangnan He

Fajie Yuan, Alexandros Karatzoglou, Ioannis Arapakis, Joemon M. Jose, and Xiangnan He. 2019. A Simple Convolutional Generative Network for Next Item Recommendation. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining, WSDM 2019, Melbourne, ...

2019
[35]

Manmatha, Mu Li, and Alexander J

Hang Zhang, Chongruo Wu, Zhongyue Zhang, Yi Zhu, Zhi Zhang, Haibin Lin, Yue Sun, Tong He, Jonas Mueller, R. Manmatha, Mu Li, and Alexander J. Smola
[36]

Xiangmin Zhou, Dong Qin, Xiaolu Lu, Lei Chen, and Yanchun Zhang. 2019. Online Social Media Recommendation Over Streams. In 35th IEEE International Conference on Data Engineering, ICDE 2019, Macao, China, April 8-11, 2019 . IEEE, 938–949

2019
[2016]

In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings

Session-based Recommendations with Recurrent Neural Networks. In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings

2016
[2017]

Enhanced LSTM for Natural Language Inference. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, ACL 2017, Vancouver, Canada, July 30 - August 4, Volume 1: Long Papers , Regina Barzilay and Min-Yen Kan (Eds.). Association for Computatio...

2017
[2019]

In Proceedings of the 28th ACM International Con- ference on Information and Knowledge Management, CIKM 2019, Beijing, China, November 3-7, 2019

BERT4Rec: Sequential Recommendation with Bidirectional Encoder Repre- sentations from Transformer. In Proceedings of the 28th ACM International Con- ference on Information and Knowledge Management, CIKM 2019, Beijing, China, November 3-7, 2019. 1441–1450

2019
[2020]

CoRR abs/2004.08955 (2020)

ResNeSt: Split-Attention Networks. CoRR abs/2004.08955 (2020)

2020 arXiv

Pith tools

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

[1] [1]

Ashton Anderson, Ravi Kumar, Andrew Tomkins, and Sergei Vassilvitskii. 2014. The dynamics of repeat consumption. In 23rd International World Wide Web Conference, WWW ’14, Seoul, Republic of Korea, April 7-11, 2014, Chin-Wan Chung, Andrei Z. Broder, Kyuseok Shim, and Torsten Suel (Eds.). ACM, 419–430

work page 2014

[2] [2]

Mozhdeh Ariannezhad, Sami Jullien, Ming Li, Min Fang, Sebastian Schelter, and Maarten de Rijke. 2022. ReCANet: A Repeat Consumption-Aware Neural Network for Next Basket Recommendation in Grocery Shopping. In SIGIR ’22: The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, Madrid, Spain, July 11 - 15, 2022 . 1240–1250

work page 2022

[3] [3]

Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2015. Neural Machine Translation by Jointly Learning to Align and Translate. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings , Yoshua Bengio and Yann LeCun (Eds.)

work page 2015

[4] [4]

Benson, Ravi Kumar, and Andrew Tomkins

Austin R. Benson, Ravi Kumar, and Andrew Tomkins. 2016. Modeling User Consumption Sequences. In Proceedings of the 25th International Conference on World Wide Web, WWW 2016, Montreal, Canada, April 11 - 15, 2016 , Jacqueline Bourdeau, Jim Hendler, Roger Nkambou, Ian Horrocks, and Ben Y. Zhao (Eds.). ACM, 519–529

work page 2016

[5] [5]

Rahul Bhagat, Srevatsan Muralidharan, Alex Lobzhanidze, and Shankar Vish- wanath. 2018. Buy It Again: Modeling Repeat Purchase Recommendations. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Dis- covery & Data Mining, KDD 2018, London, UK, August 19-23, 2018 , Yike Guo and Faisal Farooq (Eds.). ACM, 62–70

work page 2018

[6] [6]

Qian Chen, Zhen-Hua Ling, and Xiaodan Zhu. 2018. Enhancing Sentence Embed- ding with Generalized Pooling. In Proceedings of the 27th International Conference on Computational Linguistics, COLING 2018, Santa Fe, New Mexico, USA, Au- gust 20-26, 2018 , Emily M. Bender, Leon Derczynski, and Pierre Isabelle (Eds.). Association for Computational Linguistics, 1815–1826

work page 2018

[7] [7]

Qian Chen, Xiaodan Zhu, Zhen-Hua Ling, Si Wei, Hui Jiang, and Diana Inkpen

work page

[8] [8]

Xu Chen, Hongteng Xu, Yongfeng Zhang, Jiaxi Tang, Yixin Cao, Zheng Qin, and Hongyuan Zha. 2018. Sequential Recommendation with User Memory Networks. In Proceedings of the Eleventh ACM International Conference on Web Search and Data Mining, WSDM 2018, Marina Del Rey, CA, USA, February 5-9, 2018 , Yi Chang, Chengxiang Zhai, Yan Liu, and Yoelle Maarek (Eds.)...

work page 2018