pith. sign in

arxiv: 2605.19822 · v1 · pith:JK47XW7Ynew · submitted 2026-05-19 · 💻 cs.LG · cs.AI

ST-TGExplainer: Disentangling Stability and Transition Patterns for Temporal GNN Interpretability

Hongjiang Chen , Xin Zheng , Pengfei Jiao , Huan Liu , Zhidong Zhao , Huaming Wu , Feng Xia , Shirui Pan This is my paper

Pith reviewed 2026-05-20 07:07 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords temporal graph neural networksinterpretabilityexplanatory subgraphstability patternstransition patternsinformation bottleneckdisentanglement
0
0 comments X

The pith

ST-TGExplainer disentangles stability and transition patterns to explain temporal graph predictions more faithfully.

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

The paper introduces ST-TGExplainer to improve interpretability of temporal graph neural networks by separating stability patterns from repeated historical interactions and transition patterns from new first-time interactions. It trains a compact explanatory subgraph that stays predictive of the event label while using a disentangled information bottleneck to cut label-conditioned redundancy between the two pattern types. A sympathetic reader would care because prior methods ignore new interactions and therefore give incomplete accounts of why a TGNN makes its prediction in evolving systems such as social or transaction networks.

Core claim

ST-TGExplainer learns a compact explanatory subgraph that remains predictive of the event label while explicitly suppressing label-conditioned redundancy between stability and transition patterns through a disentangled information bottleneck objective, yielding explanations that account for both seen historical interactions and newly emerging first-time interactions.

What carries the argument

The disentangled information bottleneck objective, which separates stability and transition patterns while learning compact predictive explanatory subgraphs.

If this is right

  • Explanations now incorporate both repeated historical links and newly appearing links for each prediction.
  • Predictive accuracy stays high while the explanation covers a fuller set of influences.
  • Methods limited to seen interactions will show lower faithfulness on events driven by first-time interactions.
  • The same subgraph selection can be applied across different temporal graph tasks with event labels.

Where Pith is reading between the lines

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

  • The separation could be tested on streaming data to check whether transition patterns dominate during sudden network shifts.
  • Similar disentanglement might help static GNNs if stability is redefined as dense subgraphs and transitions as sparse outliers.
  • In applied settings such as recommendation systems the method could flag which new user-item edges drive a change in predicted preference.

Load-bearing premise

Stability patterns and transition patterns are sufficiently distinct that suppressing their label-conditioned redundancy produces explanations faithful to the model's actual decision process.

What would settle it

An experiment that removes the disentanglement step and measures whether faithfulness scores on new-interaction predictions fall back to the level of existing methods that consider only seen interactions.

Figures

Figures reproduced from arXiv: 2605.19822 by Feng Xia, Hongjiang Chen, Huaming Wu, Huan Liu, Pengfei Jiao, Shirui Pan, Xin Zheng, Zhidong Zhao.

Figure 1
Figure 1. Figure 1: The MRR scores of seven popular TGNNs and our proposed ST-TGEXPLAINER (‘Ours’ in red bars) for predicting seen and unseen interactions on three datasets. to effectively model newly emerging first-time interactions (i.e., unseen in history) yields unreliable explanations for novel events and behavioral shifts. To further investigate these challenges, we conduct a com￾parative analysis of seven prevalent TGN… view at source ↗
Figure 2
Figure 2. Figure 2: The overall framework of our method. as: min −I(YS; GS) − I(YT ; GT ) + β I(G t ; GS, GT ), (2) where YS and YT are auxiliary variables denoting the hy￾pothetical labels associated with the stability and transition patterns, respectively. As we can only observe the event label Y in real-world practice, both GS and GT tend to preserve information re￾lated to Y . This can cause them to encode redundant and o… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of model performance, parameter size, and training time per epoch on Enron and USLegis. that ST-TGEXPLAINER improves link prediction from both classification and ranking perspectives. In terms of AP, ST￾TGEXPLAINER achieves the best performance on five of the six datasets and remains competitive on Wikipedia, indi￾cating that the learned explanatory subgraph preserves label￾relevant information … view at source ↗
Figure 4
Figure 4. Figure 4: Ablation study results in terms of link prediction AP (%) and explanation ACC-AUC (%) [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: shows PCA projections of the learned latent vectors hS, hT , and hE on the Wikipedia and USLegis datasets. Here, hS and hT are produced by the disentanglement module, whereas hE denotes the representation learned from the explanatory graph without the stability–transition split. The separation between hS and hT suggests that ST-TGEXPLAINER captures distinct temporal patterns, in contrast to the more entang… view at source ↗
Figure 5
Figure 5. Figure 5: Hyperparameter sensitivity analysis of ST￾TGEXPLAINER on the Wikipedia dataset in terms of link pre￾diction AP (%) and explanation ACC-AUC (%). 5.5. Hyperparameter Sensitivity We conduct a sensitivity analysis on Wikipedia to examine the effects of the hyperparameters β and γ, which weight LCom and LDis, respectively, over the range [0.0, 1.0]. As shown in [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of model performance, parameter size, and training time per epoch on UCI. F.3. Ablation Study Ours w/o Dis w/o Com w/o ALL 98.4 98.7 99.0 99.3 AP (%) AP ACC 75 80 85 90 ACC-AUC (%) Wikipedia Ours w/o Dis w/o Com w/o ALL 98.1 98.4 98.7 99.0 99.3 AP (%) AP ACC 90 92 94 96 ACC-AUC (%) Reddit Ours w/o Dis w/o Com w/o ALL 96 97 98 AP (%) AP ACC 80 85 90 95 ACC-AUC (%) UCI Ours w/o Dis w/o Com w/o ALL… view at source ↗
Figure 8
Figure 8. Figure 8: Ablation study link prediction AP (%) and explanation ACC-AUC (%) values on another four datasets [PITH_FULL_IMAGE:figures/full_fig_p020_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of the embedding distributions for the stability graph (hS), transition graph (hT ), and explanatory graph without disentanglement (hE). 2854778 3462294 4031356 4249083 3462294 2709585 4249083 2709585 7220223 6312273 4031356 4249083 2709585 6312273 9788153 GNNExplainer TempME ST-TGExplainer 2829910 4785247 389070 5832309 6483033 999618 2829910 5832309 4785247 389070 389070 5832309 999618 3671… view at source ↗
read the original abstract

Temporal graph neural networks (TGNNs) have gained significant traction for solving real-world temporal graph tasks. However, their interpretability remains limited, as most TGNNs fail to identify which historical interactions most influence a given prediction. Despite promising progress on interpretable TGNNs, existing methods predominantly focus on previously seen historical interactions, which we term stability patterns, while overlooking newly emerging first-time interactions, which we term transition patterns. Both types of patterns are essential for faithful temporal explanations. To address this limitation, we propose ST-TGExplainer, a self-explainable TGNN that disentangles Stability and Transition patterns in temporal graphs for a more faithful Temporal GNN Explainer. Guided by a disentangled information bottleneck objective, ST-TGExplainer learns a compact explanatory subgraph that remains predictive of the event label while explicitly suppressing label-conditioned redundancy between stability and transition patterns. Extensive experiments demonstrate that ST-TGExplainer achieves strong predictive performance and yields more faithful explanations. Code is available at https://github.com/hjchen-hdu/ST-TGExplainer.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript introduces ST-TGExplainer, a self-explainable temporal graph neural network that disentangles stability patterns (previously seen historical interactions) and transition patterns (newly emerging first-time interactions) via a disentangled information bottleneck objective. The model learns a compact explanatory subgraph that remains predictive of the event label while suppressing label-conditioned redundancy between the two pattern types, with claims of competitive predictive performance and improved explanation faithfulness over prior TGNN interpretability methods.

Significance. If the central claims are substantiated, the work would meaningfully advance TGNN interpretability by explicitly addressing both stability and transition patterns, which the abstract correctly identifies as jointly necessary for faithful temporal explanations. The public code release supports reproducibility and is a clear strength.

major comments (2)
  1. [Experimental evaluation and faithfulness metrics] The faithfulness claim—that the disentangled subgraph recovers features actually used by the underlying TGNN rather than any label-predictive split of seen vs. new edges—rests on predictive performance of the explanatory subgraph alone. No direct verification (edge masking on a frozen TGNN, comparison to gradient attributions, or internal activation alignment) is reported, leaving open the possibility that the objective merely partitions the input into two complementary predictive pieces.
  2. [Disentangled information bottleneck objective] The disentangled information bottleneck is asserted to suppress label-conditioned redundancy between stability and transition patterns in a manner that improves faithfulness. Without ablations that isolate the redundancy term's contribution to explanation quality (as opposed to prediction accuracy) or quantitative checks that the resulting masks align with the TGNN's decision process, the mechanism's necessity for the interpretability goal remains unverified.
minor comments (2)
  1. [Abstract] The abstract states that 'extensive experiments demonstrate... more faithful explanations' but does not preview the specific faithfulness metrics or baselines used; adding one sentence would improve clarity for readers.
  2. [Methods] Notation for stability and transition masks should be introduced with explicit definitions early in the methods to avoid ambiguity when the objective is later defined.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below, clarifying aspects of our self-explainable design while committing to additional experiments that will strengthen the validation of faithfulness and the role of the disentangled objective.

read point-by-point responses

  1. Referee: [Experimental evaluation and faithfulness metrics] The faithfulness claim—that the disentangled subgraph recovers features actually used by the underlying TGNN rather than any label-predictive split of seen vs. new edges—rests on predictive performance of the explanatory subgraph alone. No direct verification (edge masking on a frozen TGNN, comparison to gradient attributions, or internal activation alignment) is reported, leaving open the possibility that the objective merely partitions the input into two complementary predictive pieces.

    Authors: We appreciate this observation on evaluation rigor. ST-TGExplainer is formulated as a self-explainable TGNN in which the explanation mechanism and prediction task are trained jointly end-to-end; there is therefore no separate frozen underlying TGNN available for independent edge masking. The predictive performance of the learned explanatory subgraph serves as a direct faithfulness measure because the model is explicitly optimized to rely on the selected stability and transition patterns for its output. This approach follows standard practice for self-explainable graph models. To address the referee’s concern and provide complementary evidence, we will add (i) edge-masking experiments that measure output change when non-explanatory edges are removed and (ii) comparisons against gradient-based attributions in the revised manuscript. revision: yes

  2. Referee: [Disentangled information bottleneck objective] The disentangled information bottleneck is asserted to suppress label-conditioned redundancy between stability and transition patterns in a manner that improves faithfulness. Without ablations that isolate the redundancy term's contribution to explanation quality (as opposed to prediction accuracy) or quantitative checks that the resulting masks align with the TGNN's decision process, the mechanism's necessity for the interpretability goal remains unverified.

    Authors: We agree that isolating the redundancy-suppression term’s specific contribution to explanation quality (distinct from its effect on accuracy) would strengthen the case for the disentangled information bottleneck. Our current experiments demonstrate overall gains in both predictive performance and explanation faithfulness when the full objective is used, but we did not report ablations that remove only the redundancy term while holding other components fixed. In the revision we will include such targeted ablations, reporting changes in standard faithfulness metrics (e.g., fidelity and sparsity) with and without the redundancy term, thereby clarifying its necessity for the interpretability objective. revision: yes

Circularity Check

0 steps flagged

No circularity: new disentangled IB objective is constructive, not a reduction to fitted inputs or self-citations

full rationale

The paper defines ST-TGExplainer via a novel disentangled information bottleneck objective that explicitly separates stability and transition patterns while preserving label predictivity. This is a self-contained modeling choice with its own loss terms rather than any derivation that reduces by construction to previously fitted quantities or prior results. No equations equate the output subgraph to an input fit, no uniqueness theorem is imported from the same authors, and no ansatz is smuggled via self-citation. The central claim rests on the new objective plus experimental validation, which is independent of the target explanations themselves. This is the normal case of a method paper introducing a fresh training criterion.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on the domain assumption that temporal graphs contain two separable classes of explanatory patterns whose redundancy can be suppressed without harming predictive power; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Stability patterns and transition patterns are distinct and both necessary for faithful temporal explanations.
    Stated directly in the abstract as the motivation for the new method.

pith-pipeline@v0.9.0 · 5741 in / 1217 out tokens · 27263 ms · 2026-05-20T07:07:34.580701+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

220 extracted references · 220 canonical work pages · 1 internal anchor

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

    work page

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

    work page

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    work page 1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

    work page

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    work page doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

    work page

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

    work page

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

    work page

  9. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

    work page

  10. [10]

    2017 , eprint=

    Attention Is All You Need , author=. 2017 , eprint=

    work page 2017

  11. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

    work page

  12. [12]

    International Conference on Learning Representations , year=

    Categorical Reparameterization with Gumbel-Softmax , author=. International Conference on Learning Representations , year=

    work page

  13. [13]

    International Conference on Learning Representations , year=

    Do We Really Need Complicated Model Architectures For Temporal Networks? , author=. International Conference on Learning Representations , year=

    work page

  14. [14]

    arXiv preprint arXiv:2302.07491 , year=

    Self-supervised temporal graph learning with temporal and structural intensity alignment , author=. arXiv preprint arXiv:2302.07491 , year=

    work page arXiv

  15. [15]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Learnable Spectral Wavelets on Dynamic Graphs to Capture Global Interactions , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  16. [16]

    Advances in Neural Information Processing Systems , volume=

    Neural temporal walks: Motif-aware representation learning on continuous-time dynamic graphs , author=. Advances in Neural Information Processing Systems , volume=

    work page

  17. [17]

    Advances in Neural Information Processing Systems , volume=

    Towards better dynamic graph learning: New architecture and unified library , author=. Advances in Neural Information Processing Systems , volume=

    work page

  18. [18]

    2010 , organization=

    Web-scale bayesian click-through rate prediction for sponsored search advertising in microsoft's bing search engine , author=. 2010 , organization=

    work page 2010

  19. [19]

    Proceedings of the eighth international workshop on data mining for online advertising , pages=

    Practical lessons from predicting clicks on ads at facebook , author=. Proceedings of the eighth international workshop on data mining for online advertising , pages=

    work page

  20. [20]

    Nature Human Behaviour , volume=

    Evolutionary dynamics of higher-order interactions in social networks , author=. Nature Human Behaviour , volume=

    work page

  21. [21]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    Dynamic graph neural networks for sequential recommendation , author=. IEEE Transactions on Knowledge and Data Engineering , volume=

    work page

  22. [22]

    International conference on machine learning , pages=

    Learning to simulate complex physics with graph networks , author=. International conference on machine learning , pages=. 2020 , organization=

    work page 2020

  23. [23]

    Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

    node2vec: Scalable feature learning for networks , author=. Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

    work page

  24. [24]

    International Conference on Learning Representations , year=

    Semi-supervised classification with graph convolutional networks , author=. International Conference on Learning Representations , year=

    work page

  25. [25]

    Transactions on Machine Learning Research , issn=

    Graph Neural Networks for Temporal Graphs: State of the Art, Open Challenges, and Opportunities , author=. Transactions on Machine Learning Research , issn=

    work page

  26. [26]

    ACM Computing Surveys (CSUR) , volume=

    A survey on embedding dynamic graphs , author=. ACM Computing Surveys (CSUR) , volume=

    work page

  27. [27]

    Proceedings of the AAAI conference on artificial intelligence , year=

    Evolvegcn: Evolving graph convolutional networks for dynamic graphs , author=. Proceedings of the AAAI conference on artificial intelligence , year=

    work page

  28. [28]

    Proceedings of the 13th international conference on web search and data mining , pages=

    Dysat: Deep neural representation learning on dynamic graphs via self-attention networks , author=. Proceedings of the 13th international conference on web search and data mining , pages=

    work page

  29. [29]

    Proceedings of the 30th ACM International Conference on Information & Knowledge Management , pages=

    Self-supervised representation learning on dynamic graphs , author=. Proceedings of the 30th ACM International Conference on Information & Knowledge Management , pages=

    work page

  30. [30]

    International Conference on Machine Learning , pages=

    On the equivalence between temporal and static equivariant graph representations , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    work page 2022

  31. [31]

    Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    Predicting dynamic embedding trajectory in temporal interaction networks , author=. Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    work page

  32. [32]

    IEEE Signal Processing Magazine , volume=

    Geometric deep learning: going beyond euclidean data , author=. IEEE Signal Processing Magazine , volume=

    work page

  33. [33]

    International Conference on Learning Representations , year=

    Graph Attention Networks , author=. International Conference on Learning Representations , year=

    work page

  34. [34]

    Advances in neural information processing systems , volume=

    Inductive representation learning on large graphs , author=. Advances in neural information processing systems , volume=

    work page

  35. [35]

    NIPS Workshop on Bayesian Deep Learning , year=

    Variational graph auto-encoders , author=. NIPS Workshop on Bayesian Deep Learning , year=

    work page

  36. [36]

    28th International Joint Conference on Artificial Intelligence, IJCAI 2019 , pages=

    Node embedding over temporal graphs , author=. 28th International Joint Conference on Artificial Intelligence, IJCAI 2019 , pages=. 2019 , organization=

    work page 2019

  37. [37]

    International Conference on Learning Representations , year=

    Inductive representation learning on temporal graphs , author=. International Conference on Learning Representations , year=

    work page

  38. [38]

    ICML 2020 Workshop on Graph Representation Learning , year=

    Temporal graph networks for deep learning on dynamic graphs , author=. ICML 2020 Workshop on Graph Representation Learning , year=

    work page 2020

  39. [39]

    Advances in Neural Information Processing Systems , volume=

    Towards better evaluation for dynamic link prediction , author=. Advances in Neural Information Processing Systems , volume=

    work page

  40. [40]

    International Conference on Learning Representations , year=

    Inductive Representation Learning in Temporal Networks via Causal Anonymous Walks , author=. International Conference on Learning Representations , year=

    work page

  41. [41]

    Advances in neural information processing systems , volume=

    Mlp-mixer: An all-mlp architecture for vision , author=. Advances in neural information processing systems , volume=

    work page

  42. [42]

    International Conference on Learning Representations , year=

    Dyrep: Learning representations over dynamic graphs , author=. International Conference on Learning Representations , year=

    work page

  43. [43]

    arXiv preprint arXiv:2105.07944 , year=

    Tcl: Transformer-based dynamic graph modelling via contrastive learning , author=. arXiv preprint arXiv:2105.07944 , year=

    work page arXiv

  44. [44]

    American journal of sociology , volume=

    Threshold models of collective behavior , author=. American journal of sociology , volume=

    work page

  45. [45]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    Influence maximization on social graphs: A survey , author=. IEEE Transactions on Knowledge and Data Engineering , volume=

    work page

  46. [46]

    Physics reports , volume=

    Temporal networks , author=. Physics reports , volume=

    work page

  47. [47]

    30th International Joint Conference on Artificial Intelligence, IJCAI 2021 , pages=

    Graph learning based recommender systems: a review , author=. 30th International Joint Conference on Artificial Intelligence, IJCAI 2021 , pages=. 2021 , organization=

    work page 2021

  48. [48]

    Journal of Network and Computer Applications , volume=

    Applications of link prediction in social networks: A review , author=. Journal of Network and Computer Applications , volume=

    work page

  49. [49]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    Temporal link prediction with motifs for social networks , author=. IEEE Transactions on Knowledge and Data Engineering , volume=

    work page

  50. [50]

    Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    Predicting path failure in time-evolving graphs , author=. Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    work page

  51. [51]

    Proceedings of the ACM Web Conference 2022 , pages=

    Trend: Temporal event and node dynamics for graph representation learning , author=. Proceedings of the ACM Web Conference 2022 , pages=

    work page 2022

  52. [52]

    Proceedings of the 2021 international conference on management of data , pages=

    Apan: Asynchronous propagation attention network for real-time temporal graph embedding , author=. Proceedings of the 2021 international conference on management of data , pages=

    work page 2021

  53. [53]

    Proceedings of the AAAI Conference on Artificial Intelligence , year=

    Ftm: A frame-level timeline modeling method for temporal graph representation learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , year=

    work page

  54. [54]

    Tgl: A general framework for temporal gnn training on billion-scale graphs , author=. Proc. VLDB Endow. , volume =

    work page

  55. [55]

    Advances in Neural Information Processing Systems , volume=

    Adaptive data augmentation on temporal graphs , author=. Advances in Neural Information Processing Systems , volume=

    work page

  56. [56]

    Advances in Neural Information Processing Systems , volume=

    Provably expressive temporal graph networks , author=. Advances in Neural Information Processing Systems , volume=

    work page

  57. [57]

    Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages=

    Learning Phrase Representations using RNN Encoder--Decoder for Statistical Machine Translation , author=. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages=. 2014 , organization=

    work page 2014

  58. [58]

    The Journal of Machine Learning Research , volume=

    Representation learning for dynamic graphs: A survey , author=. The Journal of Machine Learning Research , volume=

    work page

  59. [59]

    2023 , journal =

    Chen, Ke-Jia and Liu, Linsong and Jiang, Linpu and Chen, Jingqiang , title =. 2023 , journal =

    work page 2023

  60. [60]

    , author=

    Deep graph infomax. , author=. International Conference on Learning Representations , year=

    work page

  61. [61]

    ICML Workshop on Graph Representation Learning and Beyond , year =

    Zhu, Yanqiao and Xu, Yichen and Yu, Feng and Liu, Qiang and Wu, Shu and Wang, Liang , title =. ICML Workshop on Graph Representation Learning and Beyond , year =

    work page

  62. [62]

    Information sciences institute technical report, University of Southern California , volume=

    The Enron email dataset database schema and brief statistical report , author=. Information sciences institute technical report, University of Southern California , volume=

    work page

  63. [63]

    Journal of the American Society for Information Science and Technology , volume=

    Patterns and dynamics of users' behavior and interaction: Network analysis of an online community , author=. Journal of the American Society for Information Science and Technology , volume=

    work page

  64. [64]

    Scientific Data , volume=

    Interaction data from the copenhagen networks study , author=. Scientific Data , volume=

    work page

  65. [65]

    International Conference on Learning Representations , year=

    Learning deep representations by mutual information estimation and maximization , author=. International Conference on Learning Representations , year=

    work page

  66. [66]

    Proceedings of the ACM Web Conference 2023 , pages=

    An Attentional Multi-scale Co-evolving Model for Dynamic Link Prediction , author=. Proceedings of the ACM Web Conference 2023 , pages=

    work page 2023

  67. [67]

    Joint European Conference on Machine Learning and Knowledge Discovery in Databases , pages=

    Temporal Graph Representation Learning with Adaptive Augmentation Contrastive , author=. Joint European Conference on Machine Learning and Knowledge Discovery in Databases , pages=. 2023 , organization=

    work page 2023

  68. [68]

    Foundations and Modeling of Dynamic Networks Using Dynamic Graph Neural Networks: A Survey , year=

    Skarding, Joakim and Gabrys, Bogdan and Musial, Katarzyna , journal=. Foundations and Modeling of Dynamic Networks Using Dynamic Graph Neural Networks: A Survey , year=

    work page

  69. [69]

    2022 , booktitle =

    Wu, Lingfei and Cui, Peng and Pei, Jian and Zhao, Liang and Guo, Xiaojie , title =. 2022 , booktitle =

    work page 2022

  70. [70]

    HB-DSBM: Modeling the Dynamic Complex Networks From Community Level to Node Level , year=

    Jiao, Pengfei and Li, Tianpeng and Wu, Huaming and Wang, Chang-Dong and He, Dongxiao and Wang, Wenjun , journal=. HB-DSBM: Modeling the Dynamic Complex Networks From Community Level to Node Level , year=

    work page

  71. [71]

    International Journal of Modern Physics B , volume=

    Identifying influential nodes in dynamic social networks based on degree-corrected stochastic block model , author=. International Journal of Modern Physics B , volume=

    work page

  72. [72]

    IEEE Transactions on Knowledge and Data Engineering , year=

    Generative evolutionary anomaly detection in dynamic networks , author=. IEEE Transactions on Knowledge and Data Engineering , year=

    work page

  73. [73]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    A survey of community detection approaches: From statistical modeling to deep learning , author=. IEEE Transactions on Knowledge and Data Engineering , volume=

    work page

  74. [74]

    Proceedings of the 31st ACM International Conference on Information & Knowledge Management , pages=

    Embedding global and local influences for dynamic graphs , author=. Proceedings of the 31st ACM International Conference on Information & Knowledge Management , pages=

    work page

  75. [75]

    Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    Using Motif Transitions for Temporal Graph Generation , author=. Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    work page

  76. [76]

    Proceedings of the 28th ACM international conference on information and knowledge management , pages=

    Temporal network embedding with micro-and macro-dynamics , author=. Proceedings of the 28th ACM international conference on information and knowledge management , pages=

    work page

  77. [77]

    Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , pages=

    Discrete-time temporal network embedding via implicit hierarchical learning in hyperbolic space , author=. Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , pages=

    work page

  78. [78]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    Glodyne: Global topology preserving dynamic network embedding , author=. IEEE Transactions on Knowledge and Data Engineering , volume=

    work page

  79. [79]

    A Novel Representation Learning for Dynamic Graphs Based on Graph Convolutional Networks , year=

    Gao, Chao and Zhu, Junyou and Zhang, Fan and Wang, Zhen and Li, Xuelong , journal=. A Novel Representation Learning for Dynamic Graphs Based on Graph Convolutional Networks , year=

    work page

  80. [80]

    Temporal Aggregation and Propagation Graph Neural Networks for Dynamic Representation , year=

    Zheng, Tongya and Wang, Xinchao and Feng, Zunlei and Song, Jie and Hao, Yunzhi and Song, Mingli and Wang, Xingen and Wang, Xinyu and Chen, Chun , journal=. Temporal Aggregation and Propagation Graph Neural Networks for Dynamic Representation , year=

    work page

Showing first 80 references.