Pith. sign in

REVIEW 2 major objections 5 minor 2 cited by

TGB-Seq Benchmark: Challenging Temporal GNNs with Complex Sequential Dynamics

T0 review · 2 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read A new benchmark shows that temporal GNNs fail to learn even simple sequential dynamics, because their memory and aggregation modules cannot distinguish order-based patterns, and that this failure is masked by datasets full of repeated…

desk verdict TGB-Seq is a genuinely useful benchmark, but the 'inherently incapable' claim rests on a toy experiment that withholds node identity. read the letter →

arxiv 2502.02975 v3 pith:6VNWVF7N submitted 2025-02-05 cs.LG cs.AI

Lu Yi , Jie Peng , Yanping Zheng , Fengran Mo , Zhewei Wei , Yuhang Ye , Yue Zixuan , Zengfeng Huang This is my paper · ORCID
classification cs.LGcs.AI
keywords temporalgraphneuralnetworksfuturelinkpredictionsequentialdynamicsbenchmarkunseenedgesrepeatratiorecommendersystemsTGB-Seq
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that current temporal graph neural networks do not actually learn sequential dynamics: they can memorize repeated edges, but they cannot learn a pattern as simple as "a user who followed OpenAI and Anthropic is likely to follow AI at Meta next." Using a deliberately minimal toy graph in which the only signal is the order of item interactions and all events happen at identical timestamps, the authors find that all tested methods except a sequential-recommendation baseline score near chance, around 50-53% average precision. The paper then introduces TGB-Seq, eight large real-world datasets curated so that edges almost never repeat, forcing models to predict links they have never seen. On TGB-Seq, existing methods' performance drops substantially, several methods time out during training, and no single method wins across all domains; this is presented as evidence that temporal GNNs need to model sequence order rather than repetition.

What carries the argument

The load-bearing object is a toy temporal graph (Figure 3) built from two user groups, $u$ and $v$, and two item sequences, $\{i_0,\dots,i_4\}$ and $\{i_5,\dots,i_9\}$; every $u$-user interacts with items $\{i_0,\dots,i_3\}$ in order and every $v$-user with $\{i_5,\dots,i_8\}$, with both groups acting at identical timestamps, and the test asks whether a model can choose $i_4$ (for $u$) over $i_9$ after seeing only the four prior interactions. Because nodes and edges have no features, the only usable signal is the order of the sequence, which isolates sequential dynamics from any content cue. The second mechanism is the repeat ratio $r = |E_{\text{seen}}|/|E|$, the fraction of edges that duplicate an earlier edge; TGB-Seq's curation targets $r = 0$ (or near-zero for Yelp and Taobao) to prevent the memorization shortcut that dominates existing benchmarks.

What would settle it

Re-run the toy example of Section 3.2 with a one-hot feature that identifies each item ($i_0,\dots,i_9$) while keeping timestamps and neighborhoods identical. If any of the tested temporal GNNs then ranks $i_4$ above $i_9$ with AP well above 50%, the claim that the memory and aggregation modules are inherently incapable of learning this sequential pattern fails, because the limitation would be traceable to missing input features rather than to the modules' inability to represent order.

Watch

Extended reading notes

Core claim

The paper's central claim is that the architectural components shared by existing temporal GNNs—the memory module and the neighborhood-aggregation module—are structurally blind to sequential order. In the toy construction, items $i_4$ and $i_9$ are each connected to a mirrored group of users who interact at identical timestamps, so memory states and one-hop embeddings for the two candidates are indistinguishable; correlation-based modules such as common-neighbor counting and anonymous walks also fail because the neighborhoods mirror each other. Extending CAWN's walk length from one to two hops lifts AP only from 50.00% to 52.80% while consuming much more computation, and a three-hop walk exhausts an 80 GB GPU. The paper concludes that because existing datasets such as Wikipedia and Reddit contain 88-99% repeated edges, strong benchmark scores mostly reflect repetition-memorization, and it offers TGB-Seq—with repeat ratios of 0% on six of eight datasets—as a test that isolates whether a model can generalize to unseen edges.

Load-bearing premise

The load-bearing premise is that the toy example, in which the only signal distinguishing the two candidate items is the order of a user's past interactions and no node or edge features exist, faithfully represents the sequential dynamics found in real-world systems; if that premise fails, the conclusion that chance-level AP scores reveal an architectural limitation rather than a task or optimization difficulty does not follow.

Editorial extensions

If this is right

  • On TGB-Seq, rankings reverse relative to Wikipedia and Reddit: memory-based JODIE and TGN lead the recommendation datasets, while CAWN and DyGFormer—the strongest methods on repeated-edge benchmarks—drop sharply, showing that benchmark choice changes conclusions about which methods work.
  • The MRR gap between repeated and unseen edges on existing datasets reaches up to eightfold, so high scores on Wikipedia, Reddit, MOOC, LastFM, Enron, and Social Evo. substantially overstate generalization ability.
  • Several methods cannot finish one training epoch within 24 hours on the larger TGB-Seq datasets (e.g., TGAT and CAWN on several datasets, DyGFormer on ML-20M), so efficiency and accuracy are not simultaneously achieved by any current method.
  • TGB-Seq ships datasets, a Python package for downloading, negative-sample generation, and evaluation, and public leaderboards, making it a ready testbed for sequence-aware temporal GNNs.

Reading between the lines

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

  • Beyond the paper, the same toy construction can serve as a unit test for any new temporal GNN: a model that cannot push AP above 50% on it should not be expected to generalize on TGB-Seq, and one that solves it could be evaluated for whether the solution transfers to the real datasets.
  • The repeat-ratio statistic likely understates difficulty: a dataset with $r=0$ but highly repetitive neighborhood substructures could still be solved by aggregation shortcuts. A useful extension would be to report a 'novel-structure ratio' or to stratify MRR by whether a test edge's neighborhoods resemble training neighborhoods.
  • The paper's findings imply that temporal-GNN evaluation should adopt random negative sampling from all nodes, as TGB-Seq does, rather than historical negatives; otherwise the negative-sampling protocol itself rewards memorization and can invert rankings.
  • A testable prediction follows from the paper's mechanism analysis: adding an order-sensitive encoding (e.g., positional or sequential embeddings) to a memory or aggregation module should close much of the gap on TGB-Seq, especially on the bipartite recommendation datasets, without changing the underlying graph model.
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.

Referee Report

2 major / 5 minor

Summary. The paper argues that existing temporal-GNN benchmarks over-represent repeated edges and therefore miss a core challenge of future link prediction: complex sequential dynamics. It presents a toy experiment in which temporal GNNs score near chance while a session-based recommender (SGNN-HN) reaches perfect AP, and it uses this result to claim that models such as GraphMixer and DyGFormer are 'inherently incapable' of learning simple sequential dynamics. The authors then introduce TGB-Seq, eight large real-world temporal graphs (four bipartite recommendation networks and four non-bipartite networks) curated to have low repeat ratios, and benchmark eight temporal GNNs plus SGNN-HN under a chronological 70/15/15 split with MRR@100. They report substantial performance drops relative to Wikipedia/Reddit, dataset-dependent rankings, and high training costs, and release a pip package, datasets, and leaderboards.

Significance. The benchmark is a useful community resource: it targets an acknowledged weakness of existing temporal graph benchmarks (repeat-heavy evaluation), provides large real-world datasets with documented preprocessing, reports standard deviations over three seeds, and ships code and a leaderboard. The toy experiment is a clean attempt to isolate sequence order from other signals. However, the paper's central interpretive claim ('inherently incapable') is not supported as stated, because the toy construction removes the only information that can solve the task, and the real benchmark lacks a quantitative demonstration that its low repeat ratio translates into sequential dynamics rather than other dataset properties.

major comments (2)
  1. [Section 3.2, Table 1] The toy experiment cannot support the 'inherently incapable' claim because it strips away the only cue needed to solve the task. In the construction, i4 and i9 (and the u and v groups) interact at identical timestamps, and the text states that nodes are represented solely by null features and interaction timestamps; the paper never states whether the temporal GNNs receive node IDs or per-node embeddings. Under that input restriction, an optimal identity-free model must assign equal score to i4 and i9, so AP≈50% is the information-theoretic ceiling rather than evidence of an architectural failure. SGNN-HN reaches 100% because it is given item identities. I ask the authors to (i) state explicitly whether node IDs/per-node embeddings were provided to the temporal GNNs; (ii) add a control condition in which the temporal GNNs receive one-hot or learnable per-node embeddings; and (iii) replace 'inherently incapable' (Abstract, Section 3.2, Section 6) with a statement scoped to the identity-free input setting.
  2. [Section 4, Table 2] The paper motivates TGB-Seq as a benchmark of 'complex sequential dynamics,' but the only reported property that operationalizes this notion is the repeat ratio, and a low repeat ratio is not evidence of sequential structure: a randomly shuffled sparse graph can also have few repeated edges. To make the central interpretation credible—that the observed MRR drops reflect the need to model ordered interaction patterns—the authors should quantify sequential dynamics in the datasets (e.g., next-item transition lift or entropy, repeat-vs-exploration statistics, temporal motif counts) or run an ablation that controls for dataset size, sparsity, temporal granularity, and feature availability. Without such evidence, the performance gaps reported in Tables 3 and 4 could be driven by these other factors.
minor comments (5)
  1. [Figure 1] The x-axis labels in Figure 1 (Wikipedia, Reddit, GoogleLocal, ML-20M) do not match the caption and text, which name Yelp and Taobao as the recommendation datasets; please reconcile the figure with the reported experiments.
  2. [Section 3.2] The sentence in Section 3.2 contains a typo: 'calcu the number of common neighbors' should read 'calculate the number of common neighbors.'
  3. [Table 2] The density entry for ML-20M is typeset as 1.49×100; it should be 1.49×10^0 (i.e., 1.49%), consistent with the 'Density (%)' column.
  4. [Section 4 vs. Appendix B.1] The statement in Section 4 that 'All TGB-Seq datasets ... exclude node and edge features' conflicts with Appendix B.1, which states that ML-20M, Yelp, and GoogleLocal contain text features; please clarify that these features are excluded by the benchmark preprocessing.
  5. [Section 4, Remark] The remark that existing proposed datasets 'still contain numerous repeated edges' should acknowledge the exception of tgbl-review (0.19% repeat ratio in Table 5), which is comparable to TGB-Seq's Yelp and Taobao.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the benchmark results are independent external evaluations; only a minor non-load-bearing self-citation (Zheng et al. 2024) is present.

full rationale

The paper's TGB-Seq benchmark is an external evaluation artifact: the eight datasets are real-world graphs processed chronologically, and the reported MRR and training-cost numbers are measurements on those datasets, not quantities derived from the paper's own assumptions. No prediction reduces to a fitted parameter, and no result is obtained by renaming a known pattern or by importing a uniqueness theorem from the authors' prior work. The only self-citation is the authors' own survey (Zheng et al. 2024), used in the introduction as a general reference on temporal GNN flexibility; it is not load-bearing for any of the benchmark's claims. The §3.2 toy experiment is a controlled construction rather than a circular derivation: the symmetric, featureless item sets make AP≈50% information-theoretically forced for models that do not receive item identity, which is a validity caveat on the 'inherently incapable' wording, but it does not amount to a derivation-equivalence circularity in the benchmark's main results. Because the central benchmark findings are self-contained and externally grounded, the circularity score is low; the minor non-load-bearing self-citation accounts for the score of 2 rather than 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new theoretical entities. The free parameters listed are evaluation and preprocessing choices that materially affect the reported results. The axioms are background assumptions about the evaluation protocol and the representativeness of the toy experiment.

free parameters (3)
  • Negative sample count k = 100
    Hand-chosen for 'more robust evaluation' (Section 3.1); the reported MRR values and conclusions depend on this protocol choice.
  • Degree filter threshold = >=3
    Chosen to mitigate cold-start effects in preprocessing (Section 4); it changes which nodes and edges enter the training and test sets.
  • Chronological split ratio = 70/15/15
    Standard split chosen by hand (Section 4); the difficulty of the benchmark depends on this ratio.
assumptions (4)
  • domain assumption MRR with random negative sampling is a valid measure of future link prediction performance
    The paper relies on this to compare models; see Section 3.1. Different negative sampling strategies could change the ranking.
  • ad hoc to paper The toy example isolates sequential dynamics from other predictive signals
    The conclusion of inherent incapability depends on this premise; see Section 3.2 and Table 1.
  • domain assumption High repeat ratios in existing benchmarks are the main reason temporal GNNs perform well on them
    This assumption motivates the entire benchmark design; see Section 1 and Figure 2.
  • ad hoc to paper SGNN-HN is a representative state-of-the-art sequential recommender baseline
    Used as the main non-TGNN comparison in Tables 3 and 4, without benchmarking other sequential models.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TGB-Seq Benchmark: Challenging Temporal GNNs with Complex Sequential Dynamics." pith.science (2026). https://pith.science/paper/6VNWVF7N

@misc{pith2026250202975,
  author       = {Pith},
  title        = {Pith review of: TGB-Seq Benchmark: Challenging Temporal GNNs with Complex Sequential Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6VNWVF7N}},
  note         = {Machine review of arXiv:2502.02975}
}
read the original abstract

Future link prediction is a fundamental challenge in various real-world dynamic systems. To address this, numerous temporal graph neural networks (temporal GNNs) and benchmark datasets have been developed. However, these datasets often feature excessive repeated edges and lack complex sequential dynamics, a key characteristic inherent in many real-world applications such as recommender systems and ``Who-To-Follow'' on social networks. This oversight has led existing methods to inadvertently downplay the importance of learning sequential dynamics, focusing primarily on predicting repeated edges. In this study, we demonstrate that existing methods, such as GraphMixer and DyGFormer, are inherently incapable of learning simple sequential dynamics, such as ``a user who has followed OpenAI and Anthropic is more likely to follow AI at Meta next.'' Motivated by this issue, we introduce the Temporal Graph Benchmark with Sequential Dynamics (TGB-Seq), a new benchmark carefully curated to minimize repeated edges, challenging models to learn sequential dynamics and generalize to unseen edges. TGB-Seq comprises large real-world datasets spanning diverse domains, including e-commerce interactions, movie ratings, business reviews, social networks, citation networks and web link networks. Benchmarking experiments reveal that current methods usually suffer significant performance degradation and incur substantial training costs on TGB-Seq, posing new challenges and opportunities for future research. TGB-Seq datasets, leaderboards, and example codes are available at https://tgb-seq.github.io/.

Figures

Figures reproduced from arXiv: 2502.02975 by the authors.

Figure 1
Figure 1. The MRR scores of three selected temporal GNNs and SGNN-HN on two existing datasets (Wikipedia, Reddit) and two recommen￾dation datasets (Yelp and Taobao). one of the state-of-the-art methods for sequen￾tial recommendation to compare with tempo￾ral GNNs. Intuitively, these recommendation datasets are comparable to existing datasets (e.g., Wikipedia and Reddit), as all represent typical dynamic systems, and thus, tem… view at source ↗
Figure 2
Figure 2. The MRR scores of eight popular temporal GNNs for predicting repeated historical edges [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Toy example of sequential dynamics in a temporal graph. The bipartite graph consists [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The average training cost per epoch of eight popular temporal GNN methods on GoogleLo [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Distribution of node degree on our TGB-Seq dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. T-GRAB: A Synthetic Diagnostic Benchmark for Learning on Temporal Graphs

    cs.LG 2025-07 conditional novelty 7.0 of 10

    T-GRAB, a set of three synthetic temporal-graph tasks, shows that no current TGNN reliably does counting, delayed cause-effect, or long-range spatio-temporal reasoning.

  2. TIDFormer: Exploiting Temporal and Interactive Dynamics Makes A Great Dynamic Graph Transformer

    cs.LG 2025-05 conditional novelty 5.0 of 10

    TIDFormer, a dynamic graph Transformer with interaction-level self-attention and three encoding modules, achieves the best or second-best average rank on seven CTDG link prediction benchmarks.

Reference graph

Works this paper leans on

28 extracted references · 17 canonical work pages · cited by 2 Pith papers

  1. [1]

    Temporal graph neural networks for social recommendation

    Ting Bai, Youjie Zhang, Bin Wu, and Jian-Yun Nie. Temporal graph neural networks for social recommendation. In 2020 IEEE International Conference on Big Data (Big Data), pp. 898–903. IEEE,

    work page 2020

  2. [5]

    Tgb 2.0: A benchmark for learning on temporal knowledge graphs and heterogeneous graphs

    Julia Gastinger, Shenyang Huang, Mikhail Galkin, Erfan Loghmani, Ali Parviz, Farimah Poursafaei, Jacob Danovitch, Emanuele Rossi, Ioannis Koutis, Heiner Stuckenschmidt, et al. Tgb 2.0: A benchmark for learning on temporal knowledge graphs and heterogeneous graphs. arXiv preprint arXiv:2406.09639,

    arXiv

  3. [11]

    ISBN 9798400705052

    Association for Computing Machinery. ISBN 9798400705052. doi: 10.1145/3640457.3688119. URL https://doi.org/10.1145/3640457.3688119. Ming Li, Mozhdeh Ariannezhad, Andrew Yates, and Maarten de Rijke. Masked and swapped sequence modeling for next novel basket recommendation in grocery shopping. In Proceedings of the 17th ACM Conference on Recommender Systems...

    arXiv

  4. [12]

    Measurement and analysis of online social networks

    12 Published as a conference paper at ICLR 2025 Alan Mislove, Massimiliano Marcon, Krishna P Gummadi, Peter Druschel, and Bobby Bhattacharjee. Measurement and analysis of online social networks. In Proceedings of the 7th ACM SIGCOMM conference on Internet measurement, pp. 29–42,

    work page 2025

  5. [13]

    Sentence-BERT: Sentence embeddings using Siamese BERT- networks

    Nils Reimers and Iryna Gurevych. Sentence-BERT: Sentence embeddings using Siamese BERT- networks. In Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan (eds.), Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pp. 3982–3992, Ho...

    work page 2019

  6. [14]

    doi: 10.18653/v1/D19-1410

    Association for Computational Linguistics. doi: 10.18653/v1/D19-1410. URL https://aclanthology.org/D19-1410/. Pengjie Ren, Zhumin Chen, Jing Li, Zhaochun Ren, Jun Ma, and Maarten de Rijke. Repeatnet: A repeat aware neural recommendation machine for session-based recommendation. In The Thirty-Third AAAI Conference on Artificial Intelligence, AAAI 2019, The...

    doi:10.18653/v1/d19-1410

  7. [15]

    URL https://doi.org/10.1609/aaai.v33i01.33014806

    doi: 10.1609/AAAI.V33I01.33014806. URL https://doi.org/10.1609/aaai.v33i01.33014806. Emanuele Rossi, Ben Chamberlain, Fabrizio Frasca, Davide Eynard, Federico Monti, and Michael Bronstein. Temporal graph networks for deep learning on dynamic graphs. InICML 2020 Workshop on Graph Representation Learning,

    doi:10.1609/aaai.v33i01.33014806

  8. [16]

    URL https://doi.org/10.1109/ACCESS.2021

    doi: 10.1109/ACCESS.2021.3082932. URL https://doi.org/10.1109/ACCESS.2021. 3082932. Komal Teru, Etienne Denis, and Will Hamilton. Inductive relation prediction by subgraph reasoning. In International Conference on Machine Learning, pp. 9448–9457. PMLR,

    arXiv 2021

Show all 28 references
  1. [17]

    Dyrep: Learning representations over dynamic graphs

    Rakshit Trivedi, Mehrdad Farajtabar, Prasenjeet Biswal, and Hongyuan Zha. Dyrep: Learning representations over dynamic graphs. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019,

    2019

  2. [18]

    Tcl: Transformer-based dynamic graph modelling via contrastive learning

    Lu Wang, Xiaofu Chang, Shuang Li, Yunfei Chu, Hui Li, Wei Zhang, Xiaofeng He, Le Song, Jingren Zhou, and Hongxia Yang. Tcl: Transformer-based dynamic graph modelling via contrastive learning. arXiv preprint arXiv:2105.07944, 2021a. Meihong Wang, Linling Qiu, and Xiaoli Wang. A...

    2025 arXiv

  3. [19]

    On the feasibility of simple transformer for dynamic graph modeling

    Yuxia Wu, Yuan Fang, and Lizi Liao. On the feasibility of simple transformer for dynamic graph modeling. In Proceedings of the ACM on Web Conference 2024, pp. 870–880,

    2024

  4. [20]

    Inductive rep- resentation learning on temporal graphs

    Da Xu, Chuanwei Ruan, Evren K ¨orpeoglu, Sushant Kumar, and Kannan Achan. Inductive rep- resentation learning on temporal graphs. In 8th International Conference on Learning Rep- resentations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020 ,

    2020

  5. [23]

    14 Published as a conference paper at ICLR 2025 Table 5: A selected list of datasets used for continuous-time temporal graph learning. Dataset Nodes (users/items) Edges Timestamps Repeat ratio(%) Density(%) Bipartite Domain ML-20M 100,785/9,646 14,494,325 9,993,250 0 1.49×100 ...

    2025

  6. [24]

    This power-law behavior indicates that while a few hub nodes have a high degree of connections, the majority of nodes possess significantly fewer connections

    Figure 5 demonstrates that all the TGB-Seq datasets follow a power-law distribution, a common characteristic of real-world networks (Barab´asi, 2013). This power-law behavior indicates that while a few hub nodes have a high degree of connections, the majority of nodes possess ...

    2013

  7. [25]

    The Patent dataset is quite special that all citations of one patent are labeled with the same timestamp, specifically the publication time of the patent

    Preprocessing of the Patent dataset. The Patent dataset is quite special that all citations of one patent are labeled with the same timestamp, specifically the publication time of the patent. Therefore, we carefully select test samples to ensure that each patent has prior cita...

    2025

  8. [27]

    It includes a novel projection operation that predicts future representation trajectories of both users and items, allowing the model to anticipate future behaviors

    uses two coupled recurrent neural networks to dynamically update the states of users and items during interactions. It includes a novel projection operation that predicts future representation trajectories of both users and items, allowing the model to anticipate future behavi...

    2025

  9. [28]

    It memorizes observed interactions and uses various strategies to update its memory

    is a memory-centric approach tailored for transductive dynamic link prediction without relying on trainable parameters. It memorizes observed interactions and uses various strategies to update its memory. EdgeBank predicts future interactions based on whether the 19 Published ...

    2025

  10. [172]

    Table 6 demonstrates the performance of existing temporal GNNs on the GoogleLocal dataset when incorporating node features or edge features

    SBERT is chosen for its ability to capture the semantic information of multilingual text, which is essential as the text in GoogleLocal is multilingual. Table 6 demonstrates the performance of existing temporal GNNs on the GoogleLocal dataset when incorporating node features o...

    2025

  11. [2013]

    The nber patent citation data file: Lessons, insights and methodological tools,

    11 Published as a conference paper at ICLR 2025 Bronwyn H Hall, Adam B Jaffe, and Manuel Trajtenberg. The nber patent citation data file: Lessons, insights and methodological tools,

    2025

  12. [2014]

    URL https://doi.org/10.3115/v1/d14-1179

    doi: 10.3115/V1/D14-1179. URL https://doi.org/10.3115/v1/d14-1179. Weilin Cong, Si Zhang, Jian Kang, Baichuan Yuan, Hao Wu, Xin Zhou, Hanghang Tong, and Mehrdad Mahdavi. Do we really need complicated model architectures for temporal networks? In The Eleventh International Conf...

    doi

  13. [2015]

    Towards better dynamic graph learning: New architecture and unified library

    Le Yu, Leilei Sun, Bowen Du, and Weifeng Lv. Towards better dynamic graph learning: New architecture and unified library. In Advances in Neural Infor- mation Processing Systems 36: Annual Conference on Neural Information Process- ing Systems 2023, NeurIPS 2023, New Orleans, LA...

    2023

  14. [2017]

    Benchtemp: A general benchmark for evaluating temporal graph neural networks

    Qiang Huang, Xin Wang, Susie Xi Rao, Zhichao Han, Zitao Zhang, Yongjun He, Quanqing Xu, Yang Zhao, Zhigao Zheng, and Jiawei Jiang. Benchtemp: A general benchmark for evaluating temporal graph neural networks. In 2024 IEEE 40th International Conference on Data Engineering (ICDE...

    2024

  15. [2019]

    URL https://doi.org/10.1145/ 3292500.3330895

    doi: 10.1145/3292500.3330895. URL https://doi.org/10.1145/ 3292500.3330895. Jiacheng Li, Jingbo Shang, and Julian J. McAuley. Uctopic: Unsupervised contrastive learning for phrase representations and topic mining. In Proceedings of the 60th Annual Meeting of the Association fo...

  16. [2020]

    Network science

    Albert-L´aszl´o Barab ´asi. Network science. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 371(1987):20120375,

    1987

  17. [2021]

    Predicting dynamic embedding trajectory in temporal interaction networks

    Srijan Kumar, Xikun Zhang, and Jure Leskovec. Predicting dynamic embedding trajectory in temporal interaction networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2019, Anchorage, AK, USA, August 4-8, 2019, pp. 1269–1278,

    2019

  18. [2022]

    URL https://doi.org/10.18653/v1/2022.acl-long.426

    doi: 10.18653/V1/2022.ACL-LONG.426. URL https://doi.org/10.18653/v1/2022.acl-long.426. Jiayu Li, Aixin Sun, Weizhi Ma, Peijie Sun, and Min Zhang. Right tool, right job: Recommenda- tion for repeat and exploration consumption in food delivery. In Proceedings of the 18th ACM Con...

    2022 doi

  19. [2023]

    Yanping Zheng, Lu Yi, and Zhewei Wei

    URL http://papers.nips.cc/paper_files/paper/2023/hash/ d611019afba70d547bd595e8a4158f55-Abstract-Conference.html. Yanping Zheng, Lu Yi, and Zhewei Wei. A survey of dynamic graph neural networks. arXiv preprint arXiv:2404.18211,

    2023 arXiv

  20. [2024]

    ISBN 9798400703713

    Association for Computing Machinery. ISBN 9798400703713. doi: 10.1145/3616855.3635755. URL https://doi.org/10.1145/3616855.3635755. Kyunghyun Cho, Bart van Merrienboer,C ¸aglar G¨ulc ¸ehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. Learning phrase re...

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.