Research map / Graph learning / Knowledge and temporal graphs
DRUM: End-To-End Differentiable Rule Mining On Knowledge Graphs
Ali Sadeghian, Mohammadreza Armandpour, Patrick Ding, Daisy Zhe Wang
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 rank 4 instead of rank 1 for rule-confidence estimation (ablation) | DRUM · Family, UMLS, and Kinship (T=2 and T=3) | MRR: Family T=2: .92 -> .94; UMLS T=2: .80 -> .81; Kinship T=2: .51 -> .60; Family T=3: .91 -> .95; UMLS T=3: .77 -> .80; Kin | Improved |
| use maximum rule length 3 instead of 2 (ablation) | DRUM · WN18RR transductive | MRR: .435 -> .486 | Improved |
| use maximum rule length 3 instead of 2 (ablation) | DRUM · FB15K-237 transductive | MRR: .250 -> .343 | Improved |
| use maximum rule length 3 instead of 2 (ablation) | DRUM · Family, UMLS, and Kinship | MRR: Family: .94 -> .95; UMLS: .81 -> .80; Kinship: .60 -> .61 | Mixed |
| use DRUM instead of Neural LP | link prediction · WN18RR transductive | MRR: .435 -> .486 | Improved |
| use DRUM instead of Neural LP | link prediction · FB15K-237 transductive | MRR: .24 -> .343 | Improved |
| use DRUM instead of Neural LP | inductive link prediction · WN18 | Hits@10: 94.49 -> 95.21 | Improved |
| use DRUM instead of Neural LP | inductive link prediction · FB15K-237 | Hits@10: 27.97 -> 29.13 | Improved |
| use DRUM instead of Neural LP | rule mining · Family human rule assessment: father, sister, and uncle | number of consecutive correct rules: father: 2 -> 5; sister: 3 -> 10; uncle: 6 -> 6 | Mixed |
| use the rank-one confidence estimator instead of the higher-rank DRUM estimator (ablation) | rule-confidence estimation · logical rule quality | incorrect high-confidence rules | Hurt |
Later papers that compared with this method
6 later accepted papers report DruM as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Hard-Constrained Graph Generation with Discrete-Projection Diffusion (ICML 2026 · better on 1 of 1 shared result)
- Towards Unified Neurosymbolic Reasoning on Knowledge Graphs (TPAMI 2026 · better on 1 of 1 shared result)
- Learning from Both Structural and Textual Knowledge for Inductive Knowledge Graph Completion (NeurIPS 2023 · better on 3 of 3 shared results)
- Neural Compositional Rule Learning for Knowledge Graph Reasoning (ICLR 2023 · better on 1 of 1 shared result)
- Logical Entity Representation in Knowledge-Graphs for Differentiable Rule Learning (ICLR 2023 · better on 1 of 1 shared result)
- Inductive Relation Prediction by Subgraph Reasoning (ICML 2020 · better on 2 of 2 shared results)
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
- Logical Entity Representation in Knowledge-Graphs for Differentiable Rule Learning (ICLR 2023 · similarity 0.88)
- Topology-Aware Correlations Between Relations for Inductive Link Prediction in Knowledge Graphs (AAAI 2021 · similarity 0.87)
- Faithful Rule Extraction for Differentiable Rule Learning Models (ICLR 2024 · similarity 0.86)
- Flock: A Knowledge Graph Foundation Model via Learning on Random Walks (ICLR 2026 · similarity 0.86)
- Linear-time rule mining under formal guarantees (NeurIPS 2026 · similarity 0.86)
- Rule-Guided Graph Neural Networks for Explainable Knowledge Graph Reasoning (AAAI 2025 · similarity 0.86)
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.
