Research map / Graph learning / Knowledge and temporal graphs
Modeling Heterogeneous Hierarchies with Relation-specific Hyperbolic Cones
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 restricted rotation (cone containment) for hierarchical relations (vs RotC) (ablation) | ConE · WN18RR KG completion, all relations | MRR: .481 -> .496 | Helped |
| add restricted rotation for hierarchical relations (ablation) | ConE · WN18RR hierarchical relations | MRR: .209 -> .231 | Helped |
| add rotation in complement space for hierarchical relations (vs ConE w/o rotation) (ablation) | ConE · WN18RR KG completion | MRR: .397 -> .496 | Helped |
| add restricted rotation for non-hierarchical relation evaluation (ablation) | ConE · WN18RR non-hierarchical relations | MRR: .936 -> .939 | No clear effect |
| overlapping vs orthogonal relation-specific subspaces (ablation) | ConE · WN18RR KG completion | MRR: .493 -> .495 | No clear effect |
| learn curvature rather than fixing c=-1 (ablation) | ConE · WN18RR KG completion | MRR: .496 -> .485 | Hurt |
| increase subspace dimension (ablation) | ConE · WN18RR | hierarchical vs KG completion | Mixed |
| increase angle loss weight w (ablation) | ConE · WN18RR | ancestor-descendant mAP / KG MRR | Mixed |
| ConE vs best baseline on ancestor-descendant, hardest setting (100% inferred pairs) | ConE · DDB14 | mAP: .594 -> .818 | Helped |
| ConE vs best baseline on ancestor-descendant, 0% inferred pairs | ConE · WN18RR | mAP: .889 -> .895 | No clear effect |
| ConE vs TuckER on non-hierarchical FB15k-237 | ConE · FB15k-237 | MRR: .358 -> .345 | No clear effect |
| ConE vs RotH on hierarchical KG completion | ConE · GO21 | MRR: .151 -> .211 | Helped |
Closest papers by meaning
- Learning Hierarchy-Aware Knowledge Graph Embeddings for Link Prediction (AAAI 2020 · similarity 0.90)
- ConE: Cone Embeddings for Multi-Hop Reasoning over Knowledge Graphs (NeurIPS 2021 · similarity 0.90)
- Hyperbolic Entailment Cones for Learning Hierarchical Embeddings (ICML 2018 · similarity 0.89)
- LENA: Locality-Expanded Neural Embedding for Knowledge Base Completion (AAAI 2019 · similarity 0.87)
- Shadow Cones: A Generalized Framework for Partial Order Embeddings (ICLR 2024 · similarity 0.87)
- KG-FIT: Knowledge Graph Fine-Tuning Upon Open-World Knowledge (NeurIPS 2024 · similarity 0.87)
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.