betaFeedback

WorksComputer Science

A Topological Encoder Decoder Framework for Temporal Graph Learning

Abstract

Most temporal graph learning methods reduce future prediction to discriminative inference over past interactions, focusing on edges or node labels for a largely persistent set of nodes. This view breaks down when new nodes and edges appear, since the future graph becomes a new object with changing size, composition, and structure. We address this gap by framing temporal graph prediction as an inverse topology problem. Instead of predicting edges directly, we first predict a multiscale topological descriptor of the future graph and then reconstruct a plausible future snapshot that realizes this descriptor under inductive constraints. This approach makes global structure and node churn explicit prediction targets and produces forecasted graphs that downstream tasks can evaluate without retraining. Across 14 temporal graph datasets, we evaluate our method, TOPOGED, on node, link, and graph property prediction and compare it against state-of-the-art temporal graph models. TOPOGED significantly improves node-level forecasting accuracy over the strongest baseline, with a macro average of 0.49 versus 0.05. It also outperforms baselines in 60% of graph-structure metric evaluations and substantially increases macro-average edge prediction metrics, from near-zero to 0.12. Our results show that topology-guided graph forecasting can predict inductive future snapshots whose structure supports multiple downstream evaluations.

Introduction

Dynamic graphs, in which nodes and edges evolve over time, are central to modeling real systems such as social networks, citation graphs, online communities, and blockchain transaction networks [50, 8]. In these settings, the object of interest is often not a single future edge or label, but the next network state itself. A risk signal for an ERC20 token[52], a stress indicator for a transaction ecosystem, or a moderation statistic for an online forum depends on the structure of the future graph: which nodes remain active, which new nodes arrive, and how connectivity reorganizes.

Most temporal graph learning methods approach this problem through discriminative prediction. They learn temporal representations from past interactions and use them to score candidate edges or node labels [45, 42, 31]. This paradigm is effective when the future graph is assumed to be an extension of the past over a largely persistent node set. However, this assumption breaks down under node churn. In many real temporal graphs, new nodes appear, old nodes disappear and reappear, and edges form between both existing and newly arriving nodes. In this regime, future prediction is no longer only a ranking problem over known node pairs; it becomes a graph reconstruction problem with changing size, composition, and topology.

We argue that this setting is more naturally addressed through inductive snapshot forecasting. Given a sequence of graph snapshots up to time tt, the goal is to predict the node set and structure for the next snapshot G^t+1\widehat{G}_{t+1}. This forecast-then-predict view aligns with deployment: downstream decisions are made before the next graph is observed, and many downstream quantities can be computed only after constructing a future graph. Thus, a temporal graph model should not only rank candidate edges, but also produce a coherent future graph on which new tasks can be evaluated.

Topology [4] is useful in this setting because it compactly describes graph structure across scales while abstracting away exact node identities. Two snapshots may differ in size and node labels, yet still share similar structural organization, such as heavy-tailed degree patterns, hub-centered motifs, or changes in edge density. Rather than predicting edges directly, we summarize each snapshot by a multiscale filtration descriptor that records how node and edge mass accumulate across filtration thresholds [4]. This descriptor captures the graph’s structural state, including its final node and edge budgets, while remaining lightweight enough to forecast over time.

This leads to our central formulation: temporal graph forecasting as an inverse topology problem. Instead of predicting individual edges, we first predict the future topological descriptor Φ^(Gt+1)\widehat{\Phi}(G_{t+1}). We then reconstruct a plausible future graph whose structure realizes this predicted descriptor while respecting inductive constraints such as node arrivals, node reappearances, and edge types involving new nodes. Because the descriptor is not injective, many non-isomorphic graphs may share the same topological signature. Our decoder resolves this ambiguity by combining global topological constraints with learned edge-formation scores, seeking a reconstruction that has low descriptor error and matches empirical interaction patterns.

We introduce TOPOGED (Topological Graph Encoder Decoder), a fully inductive encoder–decoder framework for discrete-time temporal graph forecasting. TOPOGED encodes each snapshot through a degree-based filtration descriptor and forecasts the next descriptor together with node and edge budgets. A memory module estimates which historical vertices reappear; new unlabeled vertices are instantiated according to the predicted arrival budget, and a multi-phase decoder generates edges by inductive type: old–old recurring, old–old newly formed, old–new, and new–new. The resulting forecasts are complete future snapshots and can be used directly for downstream prediction without task-specific retraining. Our contributions are as follows:

  • We formulate temporal graph learning under node churn as an inductive snapshot forecasting problem and define arrival-aware node and edge types, enabling explicit modeling of reappearing old nodes, newly arriving nodes, and edges involving new nodes.

  • We introduce a topology-guided inverse formulation that forecasts a multiscale filtration descriptor and reconstructs a graph realization from the descriptor.

  • We propose TOPOGED, a topological encoder–decoder that combines descriptor prediction, memory-based node reappearance, and multi-phase edge decoding.

  • We evaluate TOPOGED across social, web and transaction networks, showing consistent gains in structural fidelity, node-level forecasting, edge-level prediction, and downstream graph-property prediction while reducing the time costs by 95% against the best competitor.

Related Work

Temporal graph models learn evolving node and edge representations, and are commonly modeled as discrete-time dynamic graphs (DTDGs) or continuous-time dynamic graphs (CTDGs). CTDGs methods [46, 19, 38, 25, 45, 23, 40, 42, 31] model the network as a stream of time-stamped events. DTDGs [34, 32, 37, 43] represent the network as a sequence of snapshot graphs collected at discrete timestamps. These approaches provide strong predictors, but most assume a fixed node universe or persistent node identities, limiting their ability to inductively generate unseen nodes and their interactions (Table 1). We focus on DTDGs because snapshot-level modeling offers a tractable framework for forecasting, enabling batch processing over time and explicit control of structural targets (e.g., node and edge budgets under churn). In contrast, CTDG generation relies on event-level modeling and point-process assumptions that are unnecessary for snapshot forecasting.

Method(s)nunseen (oo–oo)(oo–nn), (nn–nn)Full snapshotObjective
JODIE [23]×\times✓\checkmark?×\timesIS
DyRep [40]×\times✓\checkmark✓\checkmark×\timesLT
TGAT, TGN, DyGPrompt [42, 31, 46]×\times✓\checkmark✓\checkmark×\timesLN
DySAT [32]×\times✓\checkmark✓\checkmark×\timesLP
EvolveGCN [28]×\times✓\checkmark✓\checkmark×\timesMP
HTGN [43]×\times✓\checkmark✓\checkmark×\timesLP
CTAN [9]×\times✓\checkmark?×\timesLN
HOPE [25]×\times×\times×\times×\timesNT
SLATE [18]×\times✓\checkmark×\times×\timesLP
TIGGER-I, DG-Gen [10, 14]×\times✓\checkmark✓\checkmark×\timesGG
GraphPulse, MiNT [34, 35]×\times×\times×\times×\timesGP
TOPOGED (ours)✓\checkmark✓\checkmark✓\checkmark✓\checkmarkSF

Table 1. Temporal graph forecasting capabilities. oo and nn denote previously observed and newly arriving nodes; the nn column denotes the forecast of new nodes (i.e., their count). Unseen (oo–oo) denotes previously unseen edges between observed nodes. Full snapshot denotes joint forecasting of the future node and edge sets.

Dynamic graph generation. Dynamic graph generation has been studied in both discrete-time [48, 24] and continuous-time [9, 51, 10, 14] regimes. While effective for edge-event synthesis, they do not directly yield coherent snapshot forecasts in the presence of churn (node disappearance) and are often tied to training node IDs. This leaves a gap for discrete-time snapshot-forecasting methods that can inductively predict future graphs with new nodes while preserving global structural fidelity.

Topological and spectral methods. Topological and spectral methods incorporate structural priors into graph learning to capture patterns beyond local message passing, such as cycles and global connectivity changes. SLATE [18] uses temporal Laplacian features but does not address snapshot generation or inductive churn. Persistent homology supports graph-level readouts through differentiable vectorizations and layers [12, 3], while topology-aware GNNs use rewiring and topological augmentations [6, 13]. Related methods use topology to study structural evolution and regularize graph generation [44, 29]. Our method builds directly on TopER [39], which summarizes a graph using the slope and intercept of a linear fit between node and edge counts across filtration thresholds. We omit the line-fitting step but extend its filtration-based approach to temporal graph learning through compact multiscale signatures that support snapshot forecasting and generation.

Preliminaries

Dynamic graph concepts. We study discrete-time dynamic graphs represented as a sequence of snapshots G={G1,…,GT}\mathcal{G} = \{G_1, \ldots, G_T\} with Gt=(Vt,Et)G_t = (V_t, E_t). We consider the inductive setting where the node set changes over time. For time tt, we define its first appearance time as τ(v)=min⁡{t:v∈Vt}\tau(v) = \min\{t : v \in V_t\}. At time tt, we partition the node set as Vt=Vtnew∪VtoldV_t = V_t^{\mathrm{new}} \cup V_t^{\mathrm{old}}, where Vtnew={v∈Vt:τ(v)=t}V_t^{\mathrm{new}} = \{v \in V_t : \tau(v) = t\} and Vtold={v∈Vt:τ(v)<t}V_t^{\mathrm{old}} = \{v \in V_t : \tau(v) < t\}. Nodes that disappear and later reappear are treated as old upon reappearance (see Figure 1). We decompose edges in EtE_t into four disjoint types based on the appearance of edge endpoints: old–old (oo), old–new (on), new–old (no), and new–new (nn). Among old–old edges, we distinguish between previously observed pairs and newly formed ones. Following EdgeBank [30], we refer to these as oo-bank (i.e., previously seen) and oo-nobank (i.e., not seen) edges. Formally, we

Figure 1. Temporal graph evolution and next-snapshot forecasting. Old nodes and edges have appeared in at least one earlier snapshot; new nodes and edges appear for the first time in the current snapshot. Topological encodings obtained through sublevel filtrations of the observed snapshots G0G_0, G1G_1, and G2G_2 are used to forecast G^3\widehat{G}_3.

Etoo-bank={(u,v)∈Et:u,v∈Vtold(t),(u,v)∈⋃j=1t−1Ej},Eton={(u,v)∈Et:u∈Vtold(t),v∈Vtnew(t)},E_t^{\mathrm{oo\text{-}bank}} = \{(u,v) \in E_t : u,v \in V_t^{\mathrm{old}}(t), (u,v) \in\bigcup_{j=1}^{t-1} E_j\}, \qquad E_t^{\mathrm{on}} = \{(u,v) \in E_t : u \in V_t^{\mathrm{old}}(t), v \in V_t^{\mathrm{new}}(t)\},
Etoo-nobank={(u,v)∈Et:u,v∈Vtold(t),(u,v)∉⋃j=1t−1Ej},Etnn={(u,v)∈Et:u,v∈Vtnew(t)}.E_t^{\mathrm{oo\text{-}nobank}} = \{(u,v) \in E_t : u,v \in V_t^{\mathrm{old}}(t), (u,v) \notin\bigcup_{j=1}^{t-1} E_j\}, \qquad E_t^{\mathrm{nn}} = \{(u,v) \in E_t : u,v \in V_t^{\mathrm{new}}(t)\}.

Topological filtration. Topological data analysis introduces the notion of a filtration to describe how the structure of a graph emerges across scales [7]. By tracking the birth and merging of components and higher-order features as a scale parameter varies, filtrations provide a principled multiscale view of graph organization [29, 39]. However, computing full persistent homology on large or evolving graphs is often prohibitively expensive [1], both computationally and in memory, which limits its direct use in temporal graph learning. This motivates the use of lightweight filtration-based summaries that retain multiscale structural information while remaining scalable.

Given G\mathcal{G} and a filtration function, such as node degree, f:V→Rf:\mathcal{V}\to\mathbb{R}, an induced sublevel filtration using thresholds can be formed as ϵ1<⋯<ϵn\epsilon_1<\cdots<\epsilon_n: Vi={v∈V:f(v)≤ϵi}\mathcal{V}_i=\{v\in\mathcal{V}:f(v)\le\epsilon_i\}, Ei={(u,v)∈E:u,v∈Vi}\mathcal{E}_i=\{(u,v)\in\mathcal{E}:u,v\in\mathcal{V}_i\}, Gϵ1⊆⋯⊆Gϵn=G\mathcal{G}_{\epsilon_1}\subseteq\cdots\subseteq\mathcal{G}_{\epsilon_n}=\mathcal{G}. When the filtration level is clear, we omit the subscript ii.

TopoGED: Topological Graph Encoder-Decoder Framework

Philosophy. We propose the view that the evolution of a graph can be understood through its multiscale structural shape, and that topology provides a principled way to make this shape comparable across time. In temporal graphs, snapshots may differ widely in size, node identity, and local connectivity due to churn, yet they often follow coherent structural patterns as they evolve. For example, in blockchain transaction networks [33], new nodes must transact with exchange nodes to buy coins, which creates star-shaped subgraphs around exchange nodes.

Topological representations capture these patterns by summarizing how connectivity accumulates across scales, abstracting away from specific node labels and exact graph size. In this sense, topology can provide a structural blueprint across snapshots: it preserves the graph’s organization while allowing scale- and node-specific effects to be learned downstream.

Goal and high-level design. We consider discrete-time dynamic graphs with node churn. Given snapshots up to time tt, TopoGED forecasts the next snapshot G^t+1=(Vt+1,Et+1)\widehat{\mathcal{G}}_{t+1}=(\mathcal{V}_{t+1},\mathcal{E}_{t+1}) in an inductive setting where new nodes may appear. TopoGED consists of three components: (i) a scalable multiscale topological descriptor Φ(Gt)\Phi(\mathcal{G}_t) for each snapshot, computed via TDA-style filtration sequences; (ii) a topology predictor that forecasts the next-snapshot descriptor Φ(Gt+1)\Phi(\mathcal{G}_{t+1}) together with coarse structural targets, including the edge budget ∣Et+1∣|\mathcal{E}_{t+1}|, node budget ∣Vt+1∣|\mathcal{V}_{t+1}|, and the probability of new nodes pt+1p_{t+1}; and (iii) a decoder that reconstructs G^t+1\widehat{\mathcal{G}}_{t+1} by generating edges in four stages aligned with the four edge types defined in § 3. Figure 3 provides an overview of our design.

TopoGED overview

Figure 3. TopoGED overview. Each observed snapshot Gt\mathcal{G}_t is encoded into a multiscale topological descriptor Φ(Gt)\Phi(\mathcal{G}_t) via filtration. A lightweight topology predictor forecasts the next-step descriptor Φ(Gt+1)\Phi(\mathcal{G}_{t+1}) together with coarse structural targets (node and edge budgets). Conditioned on these targets and a node-memory state, the phased decoder reconstructs the next snapshot G^t+1\widehat{\mathcal{G}}_{t+1}.

Graph Encoder

References

  1. [1]Cuneyt Gurcan Akcora, Murat Kantarcioglu, Yulia R. Gel, and Baris Coskunuzer. Reduction algorithms for persistence diagrams of networks: Coraltda and prunit. In Sanmi Koyejo, S. Mohamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022. URL http://papers.nips.cc/paper_files/paper/2022/hash/9f81a6e7081497b2d458689a4ce39fc7-Abstract-Conference.html.
  2. [2]Robert Goodell Brown. Statistical forecasting for inventory control. McGraw-Hill, 1959.DOI
  3. [3]Mathieu Carrière, Frédéric Chazal, Yuichi Ike, Théo Lacombe, Martin Royer, and Yuhei Umeda. Perslay: A neural network layer for persistence diagrams and new graph topological signatures. In Silvia Chiappa and Roberto Calandra, editors, The 23rd International Conference on Artificial Intelligence and Statistics, AISTATS 2020, 26-28 August 2020, Online (Palermo, Sicily, Italy), volume 108 of Proceedings of Machine Learning Research, pages 2786–2796. PMLR, 2020. URL http://proceedings.mlr.press/v108/carriere20a.html.arxiv.org/abs/1904.09378
  4. [4]Frédéric Chazal and Bertrand Michel. An introduction to topological data analysis: Fundamental and practical aspects for data scientists. Frontiers Artif. Intell., 4:667963, 2021. doi: 10.3389/FRAI.2021.667963. URL https://doi.org/10.3389/frai.2021.667963.
  5. [5]Jinyin Chen, Xueke Wang, and Xuanheng Xu. GC-LSTM: graph convolution embedded LSTM for dynamic network link prediction. Appl. Intell., 52(7):7513–7528, 2022. doi: 10.1007/S10489-021-02518-9. URL https://doi.org/10.1007/s10489-021-02518-9.
  6. [6]Yuzhou Chen, Baris Coskunuzer, and Yulia R. Gel. Topological relational learning on graphs. In Marc’Aurelio Ranzato, Alina Beygelzimer, Yann N. Dauphin, Percy Liang, and Jennifer Wortman Vaughan, editors, Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, pages 27029–27042, 2021. URL https://proceedings.neurips.cc/paper/2021/hash/e33f4dfd9dac68f1f3afa5779648cf82-Abstract.html.arxiv.org/abs/2110.15529
  7. [7]Tamal Krishna Dey and Yusu Wang. Computational topology for data analysis. Cambridge University Press, 2022.DOI
  8. [8]ZhengZhao Feng, Rui Wang, TianXing Wang, Mingli Song, Sai Wu, and Shuibing He. A comprehensive survey of dynamic graph neural networks: Models, frameworks, benchmarks, experiments and challenges. IEEE Trans. Knowl. Data Eng., 38(1):26–46, 2026. doi: 10.1109/TKDE.2025.3621291. URL https://doi.org/10.1109/TKDE.2025.3621291.
  9. [9]Alessio Gravina, Giulio Lovisotto, Claudio Gallicchio, Davide Bacciu, and Claas Grohnfeldt. Long range propagation on continuous-time dynamic graphs. In Ruslan Salakhutdinov, Zico Kolter, Katherine A. Heller, Adrian Weller, Nuria Oliver, Jonathan Scarlett, and Felix Berkenkamp, editors, Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21-27, 2024, volume 235 of Proceedings of Machine Learning Research, pages 16206–16225. PMLR / OpenReview.net, 2024. URL https://proceedings.mlr.press/v235/gravina24a.html.arxiv.org/abs/2406.02740
  10. [10]Shubham Gupta, Sahil Manchanda, Srikanta Bedathur, and Sayan Ranu. TIGGER: scalable generative modelling for temporal interaction graphs. In Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, Thirty-Fourth Conference on Innovative Applications of Artificial Intelligence, IAAI 2022, The Twelfth Symposium on Educational Advances in Artificial Intelligence, EAAI 2022 Virtual Event, February 22 - March 1, 2022, pages 6819–6828. AAAI Press, 2022. doi: 10.1609/AAAI.V36I6.20638. URL https://doi.org/10.1609/aaai.v36i6.20638.
  11. [11]Ehsan Hajiramezanali, Arman Hasanzadeh, Krishna R. Narayanan, Nick Duffield, Mingyuan Zhou, and Xiaoning Qian. Variational graph recurrent neural networks. In Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence d’Alché-Buc, Emily B. Fox, and Roman Garnett, editors, Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, December 8-14, 2019, Vancouver, BC, Canada, pages 10700–10710, 2019. URL https://proceedings.neurips.cc/paper/2019/hash/a6b8deb7798e7532ade2a8934477d3ce-Abstract.html.arxiv.org/abs/1908.09710
  12. [12]Christoph D. Hofer, Florian Graf, Bastian Rieck, Marc Niethammer, and Roland Kwitt. Graph filtration learning. In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, volume 119 of Proceedings of Machine Learning Research, pages 4314–4323. PMLR, 2020. URL http://proceedings.mlr.press/v119/hofer20b.html.arxiv.org/abs/1905.10996
  13. [13]Max Horn, Edward De Brouwer, Michael Moor, Yves Moreau, Bastian Rieck, and Karsten M. Borgwardt. Topological graph neural networks. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net, 2022. URL https://openreview.net/forum?id=oxxUMefWEhd.arxiv.org/abs/2102.07835
  14. [14]Ryien Hosseini, Filippo Simini, Venkatram Vishwanath, and Henry Hoffmann. A deep probabilistic framework for continuous time dynamic graph generation. In Toby Walsh, Julie Shah, and Zico Kolter, editors, Thirty-Ninth AAAI Conference on Artificial Intelligence, Thirty-Seventh Conference on Innovative Applications of Artificial Intelligence, Fifteenth Symposium on Educational Advances in Artificial Intelligence, AAAI 2025, Philadelphia, PA, USA, February 25 - March 4, 2025, pages 17249–17257. AAAI Press, 2025. doi: 10.1609/AAAI.V39I16.33896. URL https://doi.org/10.1609/aaai.v39i16.33896.
  15. [15]Shenyang Huang, Farimah Poursaeif, Jacob Danovitch, Matthias Fey, Weihua Hu, Emanuele Rossi, Jure Leskovec, Michael M. Bronstein, Guillaume Rabusseau, and Reihaneh Rab-bany. Temporal graph benchmark for machine learning on temporal graphs. In Alice Oh, Tristan Naumann, Amir Globerson, Kate Saenko, Moritz Hardt, and Sergey Levine, editors, Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023, 2023. URL http://papers.nips.cc/paper_files/paper/2023/hash/066b98e63313162f6562b35962671288-Abstract-Datasets_and_Benchmarks.html.arxiv.org/abs/2307.01026
  16. [16]Søren Johansen. Estimation and hypothesis testing of cointegration vectors in gaussian vector autoregressive models. Econometrica: Journal of the Econometric Society, pages 1551–1580, 1991.DOI
  17. [17]Rudolf Emil Kalman. A new approach to linear filtering and prediction problems. Journal of Basic Engineering, 82(1):35–45, 1960.DOI
  18. [18]Yannis Karmim, Marc Lafon, Raphaël Fournier-S’niehotta, and Nicolas Thome. Supralaplacian encoding for transformer on dynamic graphs. In Amir Globersons, Lester Mackey, Danielle Belgrave, Angela Fan, Ulrich Paquet, Jakub M. Tomczak, and Cheng Zhang, editors, Advances in Neural Information Processing Systems 37: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024, 2024. URL http://papers.nips.cc/paper_files/paper/2024/hash/1ed4723f12853cbd02aecb8160f5e0c9-Abstract-Conference.html.arxiv.org/abs/2409.17986
  19. [19]Seyed Mehran Kazemi. Dynamic graph neural networks. In Graph neural networks: Foundations, frontiers, and applications, pages 323–349. Springer, 2022.DOI
  20. [20]Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Yoshua Bengio and Yann LeCun, editors, 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings. 2015. doi: 10.48550/ARXIV.1412.6980. URL https://doi.org/10.48550/arXiv.1412.6980.
  21. [21]Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings. OpenReview.net, 2017. URL https://openreview.net/forum?id=SJU4ayYgl.arxiv.org/abs/1609.02907
  22. [22]Srijan Kumar, William L. Hamilton, Jure Leskovec, and Dan Jurafsky. Community interaction and conflict on the web. In Pierre-Antoine Champin, Fabien Gandon, Mounia Lalmas, and Panagiotis G. Ipeirotis, editors, Proceedings of the 2018 World Wide Web Conference on World Wide Web, WWW 2018, Lyon, France, April 23-27, 2018, pages 933–943. ACM, 2018. doi: 10.1145/3178876.3186141. URL https://doi.org/10.1145/3178876.3186141.
  23. [23]Srijan Kumar, Xikun Zhang, and Jure Leskovec. Predicting dynamic embedding trajectory in temporal interaction networks. In Ankur Teredesai, Vipin Kumar, Ying Li, Römer Rosales, Evamaria Terzi, and George Karypis, editors, Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2019, Anchorage, AK, USA, August 4-8, 2019, pages 1269–1278. ACM, 2019. doi: 10.1145/3292500.3330895. URL https://doi.org/10.1145/3292500.3330895.
  24. [24]Fan Li, Xiaoyang Wang, Dawei Cheng, Cong Chen, Ying Zhang, and Xuemin Lin. Efficient dynamic attributed graph generation. In 41st IEEE International Conference on Data Engineering, ICDE 2025, Hong Kong, May 19-23, 2025, pages 1415–1428. IEEE, 2025. doi: 10.1109/ICDE65448.2025.00110. URL https://doi.org/10.1109/ICDE65448.2025.00110.
  25. [25]Xiao Luo, Jingyang Yuan, Zijie Huang, Huiyu Jiang, Yifang Qin, Wei Ju, Ming Zhang, and Yizhou Sun. HOPE: high-order graph ODE for modeling interacting dynamics. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors, International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA, volume 202 of Proceedings of Machine Learning Research, pages 23124–23139. PMLR, 2023. URL https://proceedings.mlr.press/v202/luo23f.html.
  26. [26]Pietro Panzarasa, Tore Opsahl, and Kathleen M. Carley. Patterns and dynamics of users’ behavior and interaction: Network analysis of an online community. J. Assoc. Inf. Sci. Technol., 60(5): 911–932, 2009. doi: 10.1002/ASI.21015. URL https://doi.org/10.1002/asi.21015.
  27. [27]Ashwin Paranjape, Austin R. Benson, and Jure Leskovec. Motifs in temporal networks. In Maarten de Rijke, Milad Shokouhi, Andrew Tomkins, and Min Zhang, editors, Proceedings of the Tenth ACM International Conference on Web Search and Data Mining, WSDM 2017, Cambridge, United Kingdom, February 6-10, 2017, pages 601–610. ACM, 2017. doi: 10.1145/3018661.3018731. URL https://doi.org/10.1145/3018661.3018731.
  28. [28]Aldo Pareja, Giacomo Domeniconi, Jie Chen, Tengfei Ma, Toyotaro Suzumura, Hiroki Kanezashi, Tim Kaler, Tao B. Schardl, and Charles E. Leiserson. Evolvgen: Evolving graph convolutional networks for dynamic graphs. In The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on Educational Advances in Artificial Intelligence, EAAI 2020, New York, NY, USA, February 7-12, 2020, pages 5363–5370. AAAI Press, 2020. doi: 10.1609/AAAI.V34I04.5984. URL https://doi.org/10.1609/aaai.v34i04.5984.
  29. [29]Joohyuk Park, Donghyun Lee, Yujee Song, Guorong Wu, and Won Hwa Kim. Topology-aware graph diffusion model with persistent homology. In Danielle Belgrave, Cheng Zhang, Laura N. Montoya, Hsuan-Tien Lin, Razvan Pascanu, Piotr Koniusz, Marzyeh Ghassemi, Nancy Chen, Iván Vladimir Meza Ruiz, and Arturo Loaiza-Bonilla, editors, Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2025, NeurIPS 2025, San Diego, CA, USA, December 2-7, 2025 / Mexico City, Mexico, November 30 - December 5, 2025, 2025. URL http://papers.nips.cc/paper_files/paper/2025/hash/bb88dbfebcb21022d32086bed631bf c5-Abstract-Conference.htmlDOI
  30. [30]Farimah Poursafaei, Shenyang Huang, Kellin Perline, and Reihaneh Rabany. Towards better evaluation for dynamic link prediction. In Sanmi Koyejo, S. Mohamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022, 2022. URL http://papers.nips.cc/paper_files/paper/2022/hash/d49042a5d49818711c401d34172f9900-Abstract-Datasets_and_Benchmarks.html.arxiv.org/abs/2207.10128
  31. [31]Emanuele Rossi, Ben Chamberlain, Fabrizio Frasca, Davide Eynard, Federico Monti, and Michael M. Bronstein. Temporal graph networks for deep learning on dynamic graphs. CoRR, abs/2006.10637, 2020. doi: 10.48550/ARXIV.2006.10637. URL https://doi.org/10.48550/arXiv.2006.10637.
  32. [32]Aravind Sankar, Yanhong Wu, Liang Gou, Wei Zhang, and Hao Yang. Dysat: Deep neural representation learning on dynamic graphs via self-attention networks. In James Caverlee, Xia (Ben) Hu, Mounia Lalmas, and Wei Wang, editors, WSDM ’20: The Thirteenth ACM International Conference on Web Search and Data Mining, Houston, TX, USA, February 3-7, 2020, pages 519–527. ACM, 2020. doi: 10.1145/3336191.3371845. URL https://doi.org/ 10.1145/3336191.3371845.DOI
  33. [33]Kiarash Shamsi, Friedhelm Victor, Murat Kantarcioglu, Yulia R. Gel, and Cuneyt Gurcan Akkora. Chartalist: Labeled graph datasets for UTXO and account-based blockchains. In Sanmi Koyejo, S. Mohamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022, 2022. URL http://papers.nips.cc/paper_files/paper/2022/hash/e245189a86310b6667ac633dbb922d50-Abstract-Datasets_and_Benchmarks.html.
  34. [34]Kiarash Shamsi, Farimah Poursafaei, Shenyang Huang, Tran Gia Bao Ngo, Baris Coskunuzer, and Cuneyt Gurcan Akkora. Graphpulse: Topological representations for temporal graph property prediction. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024. URL https://openreview.net/forum?id=DZqic2sPTY.
  35. [35]Kiarash Shamsi, Tran Gia Bao Ngo, Razieh Shirzadkhani, Shenyang Huang, Farimah Poursafaei, Poupak Azad, Reihaneh Rabbany, Baris Coskunuzer, Guillaume Rabusseau, and Cuneyt Gurcan Akkora. Mint: Multi-network transfer benchmark for temporal graph learning. In Danielle Belgrave, Cheng Zhang, Laura N. Montoya, Hsuan-Tien Lin, Razvan Pascanu, Piotr Koniusz, Marzyeh Ghassemi, Nancy Chen, Iván Vladimir Meza Ruíz, and Arturo Loaiza-Bonilla, editors, Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2025, NeurIPS 2025, San Diego, CA, USA, December 2-7, 2025 / Mexico City, Mexico, November 30 - December 5, 2025, 2025. URL http://papers.nips.cc/paper_files/paper/2025/hash/c5548168cb7324f714365a971dfe76d1-Abstract-Datasets_and_Benchmarks_Track.html.DOI
  36. [36]Christopher A. Sims. Macroeconomics and reality. Econometrica: Journal of the Econometric Society, pages 1–48, 1980.DOI
  37. [37]Junwei Su, Difan Zou, and Chuan Wu. PRES: toward scalable memory-based dynamic graph neural networks. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024. URL https://openreview.net/forum?id=gjXor87Xfy.arxiv.org/abs/2402.04284
  38. [38]Yuxing Tian, Yiyan Qi, and Fan Guo. Freedyg: Frequency enhanced continuous-time dynamic graph model for link prediction. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024. URL https://openreview.net/forum?id=82mC5ilnM.
  39. [39]Astrit Tola, Fumnilola Mary Taiwo, Cuneyt Gurcan Akkora, and Baris Coskunuzer. Toper: Topological embeddings in graph representation learning. In Danielle Belgrave, Cheng Zhang, Laura N. Montoya, Hsuan-Tien Lin, Razvan Pascanu, Piotr Koniusz, Marzyeh Ghassemi, Nancy Chen, Iván Vladimir Meza Ruíz, and Arturo Loaiza-Bonilla, editors, Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2025, NeurIPS 2025, San Diego, CA, USA, December 2-7, 2025 / Mexico City, Mexico, November 30 - December 5, 2025, 2025. URL http://papers.nips.cc/paper_files/paper/2025/hash/8027ace571384361920665f1d1b69758-Abstract-Conference.html.arxiv.org/abs/2410.01778
  40. [40]Rakshit S. 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. OpenReview.net, 2019. URL https://openreview.net/forum?id=HyePrhR5KX.
  41. [41]Ronald J. Williams and David Zipser. A learning algorithm for continually running fully recurrent neural networks. Neural Comput., 1(2):270–280, 1989. doi: 10.1162/NECO.1989.1.2.270. URL https://doi.org/10.1162/neco.1989.1.2.270.
  42. [42]Da Xu, Chuanwei Ruan, Evren Körpeoğlu, Sushant Kumar, and Kannan Achan. Inductive representation learning on temporal graphs. In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenReview.net, 2020. URL https://openreview.net/forum?id=rJeW1yHYWh.arxiv.org/abs/2002.07962
  43. [43]Menglin Yang, Min Zhou, Marcus Kalander, Zengfeng Huang, and Irwin King. Discrete-time temporal network embedding via implicit hierarchical learning in hyperbolic space. In Feida Zhu, Beng Chin Ooi, and Chunyan Miao, editors, KDD ’21: The 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, Singapore, August 14–18, 2021, pages 1975–1985. ACM, 2021. doi: 10.1145/3447548.3467422. URL https://doi.org/10.1145/3447548.3467422.
  44. [44]Dongsheng Ye, Hao Jiang, Ying Jiang, and Hao Li. Stable distance of persistent homology for dynamic graph comparison. Knowl. Based Syst., 278:110855, 2023. doi: 10.1016/J.KNOSYS.2023.110855. URL https://doi.org/10.1016/j.knosys.2023.110855.
  45. [45]Jiaxuan You, Tianyu Du, and Jure Leskovec. ROLAND: graph learning framework for dynamic graphs. In Aidong Zhang and Huzefa Rangwala, editors, KDD ’22: The 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Washington, DC, USA, August 14–18, 2022, pages 2358–2366. ACM, 2022. doi: 10.1145/3534678.3539300. URL https://doi.org/10.1145/3534678.3539300.
  46. [46]Xingtong Yu, Zhenghao Liu, Xinming Zhang, and Yuan Fang. Node-time conditional prompt learning in dynamic graphs. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. OpenReview.net, 2025. URL https://openreview.net/forum?id=kWlFvIvIqAk.arxiv.org/abs/2405.13937
  47. [47]G. Udny Yule. On the time-correlation problem, with especial reference to the variate-difference correlation method. Journal of the Royal Statistical Society, 84(4):497–537, 1921.
  48. [48]Wenbin Zhang, Liming Zhang, Dieter Pfoser, and Liang Zhao. Disentangled dynamic graph deep generation. In Carlotta Demeniconi and Ian Davidson, editors, Proceedings of the 2021 SIAM International Conference on Data Mining, SDM 2021, Virtual Event, April 29 - May 1, 2021, pages 738–746. SIAM, 2021. doi: 10.1137/1.9781611976700.83. URL https://doi.org/10.1137/1.9781611976700.83.
  49. [49]Ling Zhao, Yujiao Song, Chao Zhang, Yu Liu, Pu Wang, Tao Lin, Min Deng, and Haifeng Li. T-GCN: A temporal graph convolutional network for traffic prediction. IEEE Trans. Intell. Transp. Syst., 21(9):3848–3858, 2020. doi: 10.1109/TITS.2019.2935152. URL https://doi.org/10.1109/TITS.2019.2935152.
  50. [50]Yanping Zheng, Lu Yi, and Zhewei Wei. A survey of dynamic graph neural networks. Frontiers Comput. Sci., 19(6):196323, 2025. doi: 10.1007/S11704-024-3853-2. URL https://doi.org/10.1007/s11704-024-3853-2.
  51. [51]Dawei Zhou, Lecheng Zheng, Jiawei Han, and Jingrui He. A data-driven graph generative model for temporal interaction networks. In Rajesh Gupta, Yan Liu, Jiliang Tang, and B. Aditya Prakash, editors, KDD ’20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, CA, USA, August 23-27, 2020, pages 401–411. ACM, 2020. doi: 10.1145/3394486.3403082. URL https://doi.org/10.1145/3394486.3403082.
  52. [52]Jason Zhu, Arijit Khan, and Cuneyt Gurcan Akcora. Data depth and core-based trend detection on blockchain transaction networks. Frontiers Blockchain, 7, 2024. doi: 10.3389/FBLOC.2024.1342956. URL https://doi.org/10.3389/fbloc.2024.1342956.

Paper details

vlnc:W_82xu75ds
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents