Rule-driven inconsistency resolution for knowledge graph generation rules

Tracking #: 1947-3160

This paper is currently under review
Pieter Heyvaert
Ben De Meester
Anastasia Dimou
Ruben Verborgh

Responsible editor: 
Guest Editors Knowledge Graphs 2018

Submission type: 
Full Paper
Knowledge graphs contain annotated descriptions of entities and their interrelations, and are often generated based on rules that state how certain data sources are semantically annotated. Inconsistencies are introduced in these graphs when ontology terms are (re)used without adhering to the restrictions defined by the ontologies, affecting the quality of the graphs. Rules and ontologies are two possible root causes for these inconsistencies. Methodologies and tools were proposed to detect and resolve these inconsistencies. However, they either require the complete knowledge graph, which is not always available in a time-constrained situation; or assume that only the rules can be refined and not the ontologies. In the past, we proposed a rule-driven methodology to detect and resolve inconsistencies without requiring the complete knowledge graph, but it only allows applying a predefined set of refinements to the rules. Therefore, we propose with this paper a rule-driven methodology, extending our previous work, that considers refinements for both rules and ontologies. In this work, we provide (i) a detailed description of our methodology and its implementation; and (ii) our findings when applying the methodology to two real-life use cases: DBpedia and DBLP. The use cases show that our methodology provides valuable insights when determining which refinements should be applied to the rules and ontologies, such as the entities that need to most attention when applying refinements, and the specific ontology terms and definitions that are involved in a lot of inconsistencies and that therefore might be problematic.
Full PDF Version: 
Under Review