Research map / Graph learning / Knowledge and temporal graphs
Neuro-Symbolic Integration for Reasoning and Learning on 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.
A doctoral consortium proposal outlining a PhD thesis on neuro-symbolic methods that inject ontological rules into knowledge graph embedding link prediction, with completed and planned contributions.
Problem
Knowledge graph completion is hard because KGs are incomplete and noisy, and pure embedding methods do not exploit symbolic ontology rules; existing knowledge-enhancement methods are limited to homogeneous graphs and node classification.
Approach
Combine knowledge graph embedding scores (interpreted as probabilities) with a differentiable logical reasoning component (e.g., Scallop) so that ontological rules can be learned jointly for link prediction, and extend knowledge-enhancement layers from homogeneous to multi-relational graphs.
Key results
- A reproducibility study of knowledge-enhanced neural networks was accepted at AAAI 2024.
- Knowledge-enhanced GNNs showed no significant improvement over standalone GNNs on graph completion (DSAA23 paper).
- Node classification on heterogeneous graphs with Wikialumni is in progress, planned for 03/2024.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add knowledge enhancement layers to a graph neural network | Graph neural network for graph completion · Graph completion | unspecified | No clear effect |
Surprising findings
- Knowledge enhancement layers gave no significant gain over standalone GNNs, since the encoded rule Class(x) and Edge(x,y) implies Class(y) is already an inductive bias of message passing.
Limitations
- Main link prediction model is still under development with no experimental results reported.
- Scalability of predicting scores for all triples as a 3D tensor is an acknowledged challenge.
- Closed-world assumption in KG embeddings conflicts with KG incompleteness.
Experimental setup
Datasets: Wikialumni
Builds on: knowledge-enhanced neural networks Scallop graph neural networks
Closest papers by meaning
- Rule-Guided Graph Neural Networks for Explainable Knowledge Graph Reasoning (AAAI 2025 · similarity 0.88)
- Improving Soft Unification with Knowledge Graph Embedding Methods (ICML 2025 · similarity 0.87)
- Deciphering Raw Data in Neuro-Symbolic Learning with Provable Guarantees (AAAI 2024 · similarity 0.87)
- Position: The Road to Generalizable Neuro-Symbolic Learning Should be Paved with Foundation Models (NeurIPS 2026 · similarity 0.85)
- Rethinking Complex Queries on Knowledge Graphs with Neural Link Predictors (ICLR 2024 · similarity 0.85)
- Systematic Relational Reasoning With Epistemic Graph Neural Networks (ICLR 2025 · similarity 0.84)
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.