Research map / Graph learning / Knowledge and temporal graphs
Generative Representation Learning on Hyper-relational Knowledge Graphs via Masked Discrete Diffusion
Jaejun Lee, Seheon Kim, Joyce Jiyoung Whang
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 |
|---|---|---|---|
| add contextual message passing (context messages) (ablation) | KREPE · WikiPeople- fact generation (scratch) | accuracy: 0.578 -> 0.855 | Improved |
| add stochastic structure sampling (ablation) | KREPE · WD50K entity prediction (all positions) | MRR: 0.296 -> 0.419 | Improved |
| use attention instead of mean pooling (ablation) | KREPE · WikiPeople- entity prediction (all positions) | MRR: 0.466 -> 0.522 | Improved |
| use shared initialization instead of individual initialization (ablation) | KREPE · WikiPeople- entity prediction (all positions) | MRR: 0.390 -> 0.522 | Improved |
| replace AO-AR diffusion loss with link prediction cross-entropy (ablation) | KREPE · WikiPeople- fact generation (scratch) | accuracy: 0.855 -> 0.002 | Hurt |
| replace AO-AR diffusion loss with link prediction cross-entropy (ablation) | KREPE · WD50K entity prediction (all positions) | MRR: 0.419 -> 0.415 | No clear effect |
| use dot product instead of cosine similarity (ablation) | KREPE · WikiPeople- entity prediction (all positions) | MRR: 0.497 -> 0.522 | Improved |
| use KREPE instead of LLM baselines for fact generation | fact generation · WikiPeople- fact generation (scratch) | accuracy: 0.369 -> 0.855 | Improved |
| use KREPE instead of MAYPL for entity prediction | link prediction · WikiPeople- entity prediction (primary) | MRR: 0.519 -> 0.519 | No clear effect |
| use KREPE instead of MAYPL for entity prediction | link prediction · WD50K entity prediction (primary) | MRR: 0.381 -> 0.389 | Improved |
| use KREPE instead of HDiff for relation prediction | link prediction · WD50K relation prediction (primary) | MRR: 0.950 -> 0.963 | Improved |
| use KREPE instead of best LLM baseline for arbitrary-masking generation | fact generation · WD50K fact generation (arbitrary masking) | accuracy: 0.604 -> 0.604 | No clear effect |
Closest papers by meaning
- Structure Is All You Need: Structural Representation Learning on Hyper-Relational Knowledge Graphs (ICML 2025 · similarity 0.89)
- KGDM: A Diffusion Model to Capture Multiple Relation Semantics for Knowledge Graph Embedding (AAAI 2024 · similarity 0.86)
- Replacing Paths with Connection-Biased Attention for Knowledge Graph Completion (AAAI 2025 · similarity 0.86)
- HyperGraphRAG: Retrieval-Augmented Generation via Hypergraph-Structured Knowledge Representation (NeurIPS 2025 · similarity 0.85)
- NestE: Modeling Nested Relational Structures for Knowledge Graph Reasoning (AAAI 2024 · similarity 0.85)
- S²DN: Learning to Denoise Unconvincing Knowledge for Inductive Knowledge Graph Completion (AAAI 2025 · similarity 0.85)
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.
