A Topological Encoder Decoder Framework for Temporal Graph Learning
Organizations: University of Central Florida Orlando, FL, USA · University of Manitoba Winnipeg, Canada · Florida State University Tallahassee, FL, USA · The University of Texas at Dallas Richardson, TX, USA
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.
Figures & tables
| Method(s) | n | unseen – | – – | Full snapshot | Objective |
| JODIE [ 23 ] | ✗ | ✓ | ? | ✗ | IS |
| DyRep [ 40 ] | ✗ | ✓ | ✓ | ✗ | LT |
| TGAT, TGN, DyGPrompt [ 42 , 31 , 46 ] | ✗ | ✓ | ✓ | ✗ | LN |
| DySAT [ 32 ] | ✗ | ✓ | ✓ | ✗ | LP |
| EvolveGCN [ 28 ] | ✗ | ✓ | ✓ | ✗ | MP |
| HTGN [ 43 ] | ✗ | ✓ | ✓ | ✗ | LP |
| Dataset | Num Graphs | Avg | Avg | Avg % | Avg % oo-bank | Avg % oo-nobank | Avg % on | Avg % nn |
| CollegeMsg | 181 | 123 | 185 | 91 | 49 | 39 | 10 | 3 |
| MathOverflow | 189 | 142 | 142 | 82 | 28 | 43 | 22 | 6 |
| Adex | 299 | 249 | 200 | 51 | 28 | 10 | 52 | 9 |
| Aeternity | 235 | 303 | 282 | 54 | 34 | 12 | 43 | 11 |
| Aion | 196 | 285 | 239 | 47 | 24 | 13 | 49 | 14 |
| Aragon | 343 | 336 | 309 | 45 | 18 | 4 | 74 | 4 |
| Dataset | Structure metric wins | Node metrics | Edge metrics | |||
| TopoGED | Best baseline | TopoGED | Best baseline | TopoGED | Best baseline | |
| CollegeMsg | 8 | 4 (EvolveGCN) | (EvolveGCN) | (multiple) | ||
| MathOverflow | 9 | 2 (multiple) | (multiple) | (multiple) | ||
| Adex | 6 | 4 (HTGN) | (GCLSTM) | (GCLSTM) | ||
| Aeternity | 9 | 2 (multiple) | (ROLAND) | (multiple) | ||
| Aion | 6 | 3 (multiple) | (GCLSTM) | (multiple) | ||
| Metric | F. Probs | F. | Oracle | TopoGED |
| Avg Node Degree | ||||
| Unique Degree Count | ||||
| Degree Centrality | ||||
| Assortativity Coefficient | ||||
| Clustering Coefficient | ||||
| Density |
| Model | Time (sec) | RAM (MB) | GPU (MB) |
| ROLAND | |||
| TGCN | |||
| GCLSTM | |||
| VGRNN | |||
| HTGN | |||
| EvolveGCN |
Appendix figures & tables21 assets
Supplementary material from the paper’s appendix.
Appendix
| Symbol | Meaning |
| Total training epochs, fixed at | |
| Number of training snapshots, using the training partition | |
| to | Test set range, using the final partition |
| Number of GCN/GAT layers in the encoder | |
| Node embedding dimensionality | |
| Predictive heads for edge types |
| Dataset | SMA | VAR | V-EWMA | SSM | VECM |
| Collegemsg | 15.10 | 23.43 | 11.64 | 27.88 | 18.94 |
| Mathoverflow | 0.28 | -1.27 | 0.41 | -3.65 | -1.76 |
| Adex | 7.41 | 60.63 | 9.84 | 11.23 | 44.21 |
| Aeternity | -1.64 | 17.88 | 0.85 | 3.42 | -0.81 |
| Aion | 0.02 | 10.08 | -0.27 | 22.31 | 13.67 |
| Aragon | 6.69 | 47.72 | 5.51 | 18.55 | 39.06 |
| Dataset | SMA | VAR | V-EWMA | SSM | VECM |
| Collegemsg | 29.06 | 41.57 | 23.65 | 46.19 | 20.99 |
| Mathoverflow | 1.11 | -1.25 | 1.23 | -3.22 | -1.38 |
| Adex | 12.25 | 71.53 | 16.57 | 14.23 | 55.26 |
| Aeternity | -2.21 | 13.14 | -0.01 | 2.02 | -2.08 |
| Aion | 0.75 | 7.79 | 0.49 | 19.71 | 10.93 |
| Aragon | 7.48 | 52.68 | 6.25 | 19.80 | 43.73 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
| Metric | HTGN | ROLAND | VGRNN | TGCN | GCLSTM | EvolveGCN | TopoGED | |
| Edge Evaluation | oo-bank Precision | |||||||
| oo-bank Recall | ||||||||
| oo-bank F1 | ||||||||
| oo-nobank Precision | ||||||||
| oo-nobank Recall | ||||||||
| oo-nobank F1 |
Temporal GNNs
Temporal graph neural networks model how node relationships, graph structure, and features change over time to support prediction on evolving networks.
See all 89 papers