Neural instance retrieval matches logic system performance in knowledge graphs
NIR: Approximate Neural Instance Retrieval for ALCIQ
Machine Learning
Summary
Finding specific examples within large concept-learning systems can be very slow, creating a bottleneck. This paper introduces NIR, a group of neural network models that estimate how well certain examples fit a concept using knowledge graph data. The researchers tested different types of neural encoders and found that all of them closely match results from a trusted logic-based system, called Pellet, especially on complete data. Among these, Transformer-based models were best when some information was missing. Their code is available for others to use.
What this means in practice
- •For knowledge graph engineers: Improve instance retrieval speed in concept-learning systems using neural embeddings that closely approximate traditional logical reasoning.
- •For ontology developers: Handle incomplete knowledge bases more robustly during instance classification by using Transformer-based neural retrievers.
Authors
N'Dah Jean Kouagou, Pritilata Saha, Caglar Demir, Axel-Cyrille Ngonga Ngomo
Abstract
Instance retrieval is the main bottleneck of concept-learning systems based on refinement operators. We present NIR, a family of neural instance retrievers for ALCIQ that output graded membership scores from knowledge graph embeddings. We compare tokenization-based encoders (GRU, LSTM, Transformer) with a Compositional encoder that mirrors DL operators via primitive sub-networks. On eight benchmarks, all variants approximate Pellet closely on complete knowledge bases, while the Transformer model is most robust under incompleteness. The source code of our approach is at https://github.com/dice-group/owlapy.