Research map / Graph learning / GNN architectures
Rethinking Link Prediction for Directed Graphs
Mingguo He, Yuhe Guo, Yu Zheng, Zhewei Wei, Stephan Günnemann, Xiaokui Xiao
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 |
|---|---|---|---|
| compare simple MLP against DUPLEX under existing setup | MLP · Cora (DUPLEX setup, test edges in training) | DP AUC: DUPLEX 96.34 vs MLP 99.25 DP AUC, but DUPLEX better on 3C ACC 92.41 vs 79.97 | Mixed |
| remove test edges from training graph (fix label leakage) | DUPLEX · Cora (DUPLEX setup) | EP ACC: 93.49 -> 87.43 | Hurt |
| remove test edges from training graph (fix label leakage) | DUPLEX · CiteSeer (DUPLEX setup) | EP ACC: 92.11 -> 83.59 | Hurt |
| add polynomial graph filter (SDGAE) in place of DiGAE convolution (ablation) | DiGAE · DirLinkBench Cora-ML | Hits@100: 82.06 -> 90.37 | Improved |
| add polynomial graph filter (SDGAE) in place of DiGAE convolution (ablation) | DiGAE · DirLinkBench CiteSeer | Hits@100: 83.64 -> 93.69 | Improved |
| add polynomial graph filter (SDGAE) in place of DiGAE convolution (ablation) | DiGAE · DirLinkBench Photo | Hits@100: 55.05 -> 68.84 | Improved |
| use symmetric normalization on DiGAE (ablation) | DiGAE · DirLinkBench Cora-ML | Hits@100: 82.06 -> 70.89 | Hurt |
| use symmetric normalization on DiGAE (ablation) | DiGAE · DirLinkBench CiteSeer | Hits@100: 83.64 -> 71.60 | Hurt |
| increase polynomial order K (ablation) | SDGAE · Cora-ML and CiteSeer | Hits@100: consistently improves with increasing K, optimal at K = 5 | Improved |
| increase number of convolution layers L (ablation) | DiGAE · Cora-ML and CiteSeer | Hits@100: performance declines as layers increase | Hurt |
| add residual connections (ablation) | DiGAE · Cora-ML and CiteSeer | Hits@100: simply adding residual connections fails to improve DiGAE's performance | No clear effect |
| use BCE loss instead of CE loss (ablation) | MagNet · DirLinkBench | Hits@100: BCE consistently outperforms CE | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- STRAP — STRAP: Spatio-Temporal Pattern Retrieval for Out-of-Distribution Generalization (NeurIPS 2025) · better on 4 of 7 shared results
- MagNet — MagNet: A Neural Network for Directed Graphs (NeurIPS 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Topological Perspective on Demystifying GNN-Based Link Prediction Performance (ICLR 2024 · similarity 0.90)
- GNNs Meet Sequence Models Along the Shortest-Path: an Expressive Method for Link Prediction (NeurIPS 2026 · similarity 0.88)
- Structural Information Enhanced Graph Representation for Link Prediction (AAAI 2024 · similarity 0.88)
- Bridging Theory and Practice in Link Representation with Graph Neural Networks (NeurIPS 2025 · similarity 0.88)
- Implicit degree bias in the link prediction task (ICML 2025 · similarity 0.88)
- Revisiting Link Prediction: a data perspective (ICLR 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.
