Research map / Graph learning / Knowledge and temporal graphs
Towards Better Dynamic Graph Learning: New Architecture and Unified Library
Le Yu, Leilei Sun, Bowen Du, Weifeng Lv
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use DyGFormer instead of the strongest baseline | Transductive dynamic link prediction across 13 datasets with random negative sampling · 13 datasets | average rank (AP): 2.54 -> 1.62 | Improved |
| Use DyGFormer instead of the strongest baseline | Transductive dynamic link prediction across 13 datasets with historical negative sampling · 13 datasets | average rank (AP): 3.85 -> 2.62 | Improved |
| Use DyGFormer instead of the strongest baseline | Transductive dynamic link prediction across 13 datasets with inductive negative sampling · 13 datasets | average rank (AP): 4.15 -> 3.23 | Improved |
| Add neighbor co-occurrence encoding | TCL · 13 datasets | performance: 5.36% | Improved |
| Add neighbor co-occurrence encoding | GraphMixer · 13 datasets | performance: 1.86% | Improved |
| Use patching rather than processing each interaction separately (ablation) | DyGFormer · LastFM | performance: 0.31% | Improved |
| Use patching rather than processing each interaction separately (ablation) | DyGFormer · Can. Parl. | performance: 0.74% | Improved |
| Add patching (ablation) | DyGFormer · Dynamic link prediction | performance | Improved |
| Add neighbor co-occurrence encoding (ablation) | DyGFormer · Dynamic link prediction | performance | Improved |
| Add time encoding (ablation) | DyGFormer · Dynamic link prediction | performance | Improved |
| Mix source-node and destination-node sequences in the Transformer (ablation) | DyGFormer · Dynamic link prediction | performance | Improved |
| Use longer historical sequences with patching (ablation) | DyGFormer · LastFM and Can. Parl. | performance | Improved |
Closest papers by meaning
- Node-Time Conditional Prompt Learning in Dynamic Graphs (ICLR 2025 · similarity 0.91)
- DyG-Mamba: Continuous State Space Modeling on Dynamic Graphs (NeurIPS 2025 · similarity 0.89)
- A Deep Probabilistic Framework for Continuous Time Dynamic Graph Generation (AAAI 2025 · similarity 0.89)
- Towards Dynamic Message Passing on Graphs (NeurIPS 2024 · similarity 0.89)
- DG-Mamba: Robust and Efficient Dynamic Graph Structure Learning with Selective State Space Models (AAAI 2025 · similarity 0.88)
- Every Node Is Different: Dynamically Fusing Self-Supervised Tasks for Attributed Graph Clustering (AAAI 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
