Wikipedia Knowledge Graph for Explainable AI

Abstract

Explainable artificial intelligence (XAI) requires domain information to explain a system’s decisions, for which structured forms of domain information like Knowledge Graphs (KGs) or ontologies are best suited. As such, readily available KGs are important to accelerate progress in XAI. To facilitate the advancement of XAI, we present the cycle-free Wikipedia Knowledge Graph (WKG) based on information from English Wikipedia. Each Wikipedia article title, its corresponding category, and the category hierarchy are transformed into different entities in the knowledge graph. Along with cycle-free version we also provide the original knowledge graph as it is. We evaluate whether the WKG is helpful to improve XAI compared with existing KGs, finding that WKG is better suited than the current state of the art. We also compare the cycle-free WKG with the Suggested Upper Merged Ontology (SUMO) and DBpedia schema KGs, finding minimal to no information loss.