[Novel] Knowledge gap discovery: A case study of Wikidata

Millenio Ramadizsa, Fariz Darari, Werner Nutt, Simon Razniewski

Abstract:

Society, science, and economy are becoming more and more data-driven, and therefore the study of gaps in knowledge gains importance. The arguably most prominent public source of structured knowledge is Wikidata, which contains impressive amounts of knowledge, but nonetheless comes with surprising gaps.

In this paper we propose a framework for identifying class-level knowledge gaps in Wikidata, based on the concepts of gap properties, i.e., properties that mostly exist for prominent entities, but are missing in the tail, and the gap property ratio. We conduct analysis for a varied set of 20 classes, and show that our framework can discover knowledge gaps, that may guide contributors towards addressing them.

Submission Number: 18