Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Reasoning for Robust Instance Retrieval in $\mathcal{SHOIQ}$

Louis Mozart Kamdem Teyou, Luke Friedrichs, N'Dah Jean Kouagou, Caglar Demir, Yasir Mahmood, Stefan Heindorf, Axel-Cyrille Ngonga Ngomo

TL;DR

The paper addresses robust instance retrieval for expressive description logics under real-world imperfections by introducing the Embedding Based Reasoner (EBR), which uses knowledge graph embeddings to approximate symbolic reasoning in $\mathcal{SHOIQ}$. EBR constructs a knowledge graph from DL axioms, trains a KGE, and defines neural semantics via a thresholded triple scoring function $\phi(x,r,y)=\sigma(\phi_\Theta(x,r,y))$, enabling continuous grounding of DL constructs. Across six diverse datasets, EBR (with ComplEx at 128 dimensions) achieves near-perfect retrieval on complete data and demonstrates superior robustness to noise and incompleteness compared with symbolic reasoners, while offering favorable runtime characteristics. The work also shows that EBR integrations support concept learning with multiple DL-Learner-based learners, highlighting practical impact for scalable, explainable, and robust knowledge-base reasoning in real-world settings, with code and data publicly available.

Abstract

Concept learning exploits background knowledge in the form of description logic axioms to learn explainable classification models from knowledge bases. Despite recent breakthroughs in neuro-symbolic concept learning, most approaches still cannot be deployed on real-world knowledge bases. This is due to their use of description logic reasoners, which are not robust against inconsistencies nor erroneous data. We address this challenge by presenting a novel neural reasoner dubbed EBR. Our reasoner relies on embeddings to approximate the results of a symbolic reasoner. We show that EBR solely requires retrieving instances for atomic concepts and existential restrictions to retrieve or approximate the set of instances of any concept in the description logic $\mathcal{SHOIQ}$. In our experiments, we compare EBR with state-of-the-art reasoners. Our results suggest that EBR is robust against missing and erroneous data in contrast to existing reasoners.

Neural Reasoning for Robust Instance Retrieval in $\mathcal{SHOIQ}$

TL;DR

The paper addresses robust instance retrieval for expressive description logics under real-world imperfections by introducing the Embedding Based Reasoner (EBR), which uses knowledge graph embeddings to approximate symbolic reasoning in . EBR constructs a knowledge graph from DL axioms, trains a KGE, and defines neural semantics via a thresholded triple scoring function , enabling continuous grounding of DL constructs. Across six diverse datasets, EBR (with ComplEx at 128 dimensions) achieves near-perfect retrieval on complete data and demonstrates superior robustness to noise and incompleteness compared with symbolic reasoners, while offering favorable runtime characteristics. The work also shows that EBR integrations support concept learning with multiple DL-Learner-based learners, highlighting practical impact for scalable, explainable, and robust knowledge-base reasoning in real-world settings, with code and data publicly available.

Abstract

Concept learning exploits background knowledge in the form of description logic axioms to learn explainable classification models from knowledge bases. Despite recent breakthroughs in neuro-symbolic concept learning, most approaches still cannot be deployed on real-world knowledge bases. This is due to their use of description logic reasoners, which are not robust against inconsistencies nor erroneous data. We address this challenge by presenting a novel neural reasoner dubbed EBR. Our reasoner relies on embeddings to approximate the results of a symbolic reasoner. We show that EBR solely requires retrieving instances for atomic concepts and existential restrictions to retrieve or approximate the set of instances of any concept in the description logic . In our experiments, we compare EBR with state-of-the-art reasoners. Our results suggest that EBR is robust against missing and erroneous data in contrast to existing reasoners.
Paper Structure (22 sections, 2 equations, 3 figures, 6 tables)

This paper contains 22 sections, 2 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background and Related Works
  3. Knowledge Bases in Description Logics
  4. Symbolic Reasoning over Knowledge Bases
  5. Knowledge Graph Embeddings
  6. Approach
  7. Knowledge Graph Embedding and Construction
  8. Prediction Mechanism
  9. Mapping DL Syntax to Neural Semantics
  10. Experimental Setup
  11. Datasets
  12. Tasks
  13. Task 1: Instance Retrieval.
  14. Task 2: Robustness to Noise and Incompleteness.
  15. Evaluation Protocol.
  16. ...and 7 more sections

Figures (3)

  • Figure 1: Ebr average performance with different embedding models and dimensions on different datasets. All models share the same legend shown once (right of the top-left panel).
  • Figure 2: Jaccard score of each reasoner across datasets and different levels of noise
  • Figure 3: Jaccard score of each reasoner across datasets and different levels of incompleteness