DyNA-Mixer: Decoupling Time and Topology for Scalable Dynamic Node Affinity Prediction
Organizations: East China Normal University, Shanghai, China · Shanghai Jiao Tong University, Shanghai, China
Abstract
Continuous-time dynamic graphs preserve the ordering, timing, and evolving strength of interactions that static snapshots discard, making dynamic graph mining essential for evolving systems such as recommender systems, trade networks, and transaction platforms. Research in this area has centered on dynamic link prediction, which forecasts only whether a future interaction will occur, whereas many applications also need its intensity. The dynamic node affinity prediction (DNAP) problem generalizes this task to continuous targets and full-candidate ranking, making it more broadly applicable but substantially harder. This raises two coupled difficulties: ranking requires a predictor expressive enough to capture both multi-scale temporal evolution and graph-wide collaborative patterns, yet running such a predictor over every source--destination pair is computationally prohibitive. We propose DyNA-Mixer, which decouples temporal forecasting from collaborative structure modeling. It adopts a multi-scale temporal predictor for pair-specific affinity evolution, constructs a Virtual Collaborative Graph that infers query-time relationships and aggregates trends from related interactions, and fuses the two through a gated residual mechanism. A sparsity-aware evaluation strategy further restricts expensive forecasting to pairs with recent non-zero affinities. Experiments on four Temporal Graph Benchmark datasets for the DNAP problem show that DyNA-Mixer achieves new state-of-the-art NDCG@10 on every dataset, including a 17.8% relative test improvement over the strongest baseline on tgbn-token, the largest dataset.
Figures & tables
| Scale | Dataset | Node Count | Temporal Edge Count | Temporal Step Count | Surprise |
|---|---|---|---|---|---|
| small | tgbn-trade | 255 | 468,245 | 32 | 0.023 |
| medium | tgbn-genre | 1,505 | 17,858,395 | 133,758 | 0.005 |
| large | tgbn-reddit | 11,766 | 27,174,118 | 21,889,537 | 0.013 |
| large | tgbn-token | 61,756 | 72,936,998 | 2,036,524 | 0.014 |
| Method | tgbn-trade | tgbn-genre | tgbn-reddit | tgbn-token | ||||
|---|---|---|---|---|---|---|---|---|
| Val | Test | Val | Test | Val | Test | Val | Test | |
| Persistent Forecast | 0.860 | 0.855 | 0.350 | 0.357 | 0.380 | 0.369 | 0.403 | 0.430 |
| Moving Avg | 0.793 | 0.777 | 0.496 | 0.497 | 0.498 | 0.480 | 0.401 | 0.414 |
| Historical Avg | 0.793 | 0.777 | 0.478 | 0.472 | 0.499 | 0.481 | 0.402 | 0.415 |
| JODIE | – | – | ||||||
| TGAT | – | – | ||||||
| Method | tgbn-trade | tgbn-genre | tgbn-reddit | tgbn-token | ||||
|---|---|---|---|---|---|---|---|---|
| Val | Test | Val | Test | Val | Test | Val | Test | |
| TimeMixer only | 0.873 | 0.863 | 0.533 | 0.542 | 0.602 | 0.589 | 0.498 | 0.516 |
| DyNA-Mixer (TimeMixer + VCG) | 0.885 | 0.871 | 0.535 | 0.544 | 0.605 | 0.591 | 0.510 | 0.524 |
| Dataset | Dense Time | Sparse-Aware Time | Dense Val | Dense Test | Sparse-Aware Val | Sparse-Aware Test |
|---|---|---|---|---|---|---|
| tgbn-trade | 6.4s | 16.9s | 0.885 | 0.870 | 0.885 | 0.871 |
| tgbn-genre | 1057.2s | 726.7s | 0.535 | 0.543 | 0.535 | 0.543 |
| tgbn-reddit | OOM | 4721.4s | – | – | 0.605 | 0.591 |
| tgbn-token | OOM | 3470.8s | – | – | 0.510 | 0.524 |
Appendix figures & tables9 assets
Supplementary material from the paper’s appendix.
Appendix
| Symbol | Description |
|---|---|
| , | Sets of source and destination nodes. |
| , | Numbers of source and destination nodes, respectively. |
| Continuous-time dynamic graph observed up to time . | |
| Interaction event with source, destination, timestamp, and edge attributes. | |
| Number of interaction events observed by time . | |
| , | Query source node and candidate destination node. |
| Dataset | History Length | TSP Params | VCG Params | Total Params |
|---|---|---|---|---|
| tgbn-trade | 3 | 21,475 | 122,737 | 144,212 |
| tgbn-trade | 22,085 | 122,961 | 145,046 | |
| tgbn-trade | 30 | 26,895 | 123,601 | 150,496 |
| tgbn-trade | 60 | 42,735 | 124,561 | 167,296 |
| Dataset | History Length | TSP Params | VCG Params | Total Params |
|---|---|---|---|---|
| tgbn-genre | 3 | 23,539 | 122,737 | 146,276 |
| tgbn-genre | 10 | 24,149 | 122,961 | 147,110 |
| tgbn-genre | 30 | 28,959 | 123,601 | 152,560 |
| tgbn-genre | 60 | 44,799 | 124,561 | 169,360 |
| tgbn-genre | 90 | 70,989 | 125,521 | 196,510 |
| tgbn-genre | 120 | 107,529 | 126,481 | 234,010 |
| Dataset | History Length | TSP Params | VCG Params | Total Params |
|---|---|---|---|---|
| tgbn-reddit | 3 | 25,019 | 122,737 | 147,756 |
| tgbn-reddit | 10 | 25,629 | 122,961 | 148,590 |
| tgbn-reddit | 30 | 30,439 | 123,601 | 154,040 |
| tgbn-reddit | 60 | 46,279 | 124,561 | 170,840 |
| tgbn-reddit | 90 | 72,469 | 125,521 | 197,990 |
| tgbn-reddit | 120 | 109,009 | 126,481 | 235,490 |
| Dataset | History Length | TSP Params | VCG Params | Total Params |
|---|---|---|---|---|
| tgbn-token | 3 | 27,443 | 122,865 | 150,308 |
| tgbn-token | 10 | 28,053 | 123,089 | 151,142 |
| tgbn-token | 30 | 32,863 | 123,729 | 156,592 |
| tgbn-token | 60 | 48,703 | 124,689 | 173,392 |
| tgbn-token | 74,893 | 125,649 | 200,542 | |
| tgbn-token | 120 | 111,433 | 126,609 | 238,042 |
Dynamic Graph Learning
Dynamic graph learning models how networks and their node features change over time to predict links, events, labels, and system behavior.
See all 137 papers