Research map / Graph learning / Knowledge and temporal graphs
Anchoring Path for Inductive Relation Prediction in Knowledge Graphs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Anchoring Path Sentence Transformer (APST), which adds filtered head/tail anchoring paths and retrieved Wikipedia/Wiktionary descriptions to a Sentence Transformer for inductive, explainable relation prediction.
Problem
Path-based relation prediction in knowledge graphs relies on closed paths connecting head and tail, which are often absent in incomplete KGs, limiting performance.
Approach
Defines Anchoring Paths (paths attached to only the head or tail entity), filters them with AP accuracy and AP recall, and encodes them with CPs via a fine-tuned all-mpnet-base-v2 Sentence Transformer that uses retrieved detailed entity descriptions.
Key results
- APST reaches MRR 0.774 on transductive FB15k-237 versus 0.720 for KRST, and MRR 0.764 on inductive FB15k-237 versus 0.716 for KRST.
- APST achieves SOTA in 30 of 36 transductive, inductive, and few-shot settings, per the abstract.
- Few-shot: APST attains MRR 0.724 on transductive FB15k-237 with 1000 triplets, versus 0.696 for KRST.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use APs with detailed descriptions instead of CPs only with detailed descriptions (ablation) | APST · WN18RR, FB15k-237, and NELL-995, transductive and inductive | MRR | Helped |
| Use APs with detailed descriptions instead of CPs only with detailed descriptions (ablation) | APST · WN18RR, FB15k-237, and NELL-995, transductive and inductive | Hit@1 | Helped |
| Add detailed descriptions to AP inputs (ablation) | APST · WN18RR, FB15k-237, and NELL-995 | MRR | Mixed |
| Add detailed descriptions to AP inputs (ablation) | APST · WN18RR, FB15k-237, and NELL-995 | Hit@1 | Mixed |
| Add APs to detailed-description CP inputs (ablation) | APST · WN18RR, FB15k-237, and NELL-995 | MRR | Helped |
| Add APs to detailed-description CP inputs (ablation) | APST · WN18RR, FB15k-237, and NELL-995 | Hit@1 | Helped |
Surprising findings
- Detailed descriptions do not always improve closed-path-only performance (better in only 5 of 12 comparisons), attributed to mismatch between Wikipedia/Wiktionary descriptions and NELL-995/WN18RR entities.
Limitations
- Current model primarily targets relatively small datasets.
- Performance on NELL-995 is limited by retrieved descriptions.
- Detailed descriptions can hurt closed-path-only performance due to mismatched sources.
Experimental setup
Datasets: WN18RR FB15k-237 NELL-995 Wikipedia descriptions Wiktionary descriptions
Benchmarks: WN18RR FB15k-237 NELL-995
Metrics: MRR Hit@1
Baselines: RuleN GRAIL MINERVA TuckER KG-BERT BERTRL KRST
Builds on: KRST BERTRL Sentence Transformers
Closest papers by meaning
- Knowledge Graph Completion with Relation-Aware Anchor Enhancement (AAAI 2025 · similarity 0.89)
- Replacing Paths with Connection-Biased Attention for Knowledge Graph Completion (AAAI 2025 · similarity 0.89)
- Context-aware Inductive Knowledge Graph Completion with Latent Type Constraints and Subgraph Reasoning (AAAI 2025 · similarity 0.88)
- Towards Global-Topology Relation Graph for Inductive Knowledge Graph Completion (AAAI 2025 · similarity 0.88)
- S²DN: Learning to Denoise Unconvincing Knowledge for Inductive Knowledge Graph Completion (AAAI 2025 · similarity 0.87)
- Reasoning of Large Language Models over Knowledge Graphs with Super-Relations (ICLR 2025 · 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.