Li, Na, Bouraoui, Zied and Schockaert, Steven ![]() |
Preview |
PDF
- Accepted Post-Print Version
Download (261kB) | Preview |
Abstract
Many methods have been proposed to automatically extend knowledge bases, but the vast majority of these methods focus on finding plausible missing facts, and knowledge graph triples in particular. In this paper, we instead focus on automatically extending ontologies that are encoded as a set of existential rules. In particular, our aim is to find rules that are plausible, but which cannot be deduced from the given ontology. To this end, we propose a graph-based representation of rule bases. Nodes of the considered graphs correspond to predicates, and they are annotated with vectors encoding our prior knowledge about the meaning of these predicates. The vectors may be obtained from external resources such as word embeddings or they could be estimated from the rule base itself. Edges connect predicates that co-occur in the same rule and their annotations reflect the types of rules in which the predicates co-occur. We then use a neural network model based on Graph Convolutional Networks (GCNs) to refine the initial vector representation of the predicates, to obtain a representation which is predictive of which rules are plausible. We present experimental results that demonstrate the strong performance of this method.
Item Type: | Conference or Workshop Item (Paper) |
---|---|
Date Type: | Publication |
Status: | Published |
Schools: | Schools > Computer Science & Informatics |
Publisher: | Springer |
ISBN: | 978-3-030-30792-9 |
Date of First Compliant Deposit: | 14 August 2019 |
Last Modified: | 11 Aug 2025 14:30 |
URI: | https://orca.cardiff.ac.uk/id/eprint/124041 |
Citation Data
Cited 8 times in Scopus. View in Scopus. Powered By Scopus® Data
Actions (repository staff only)
![]() |
Edit Item |