Towards Automated Technologies in the Referencing Quality of Wikidata

Abstract

Wikidata is a general-purpose knowledge graph with the content being crowd-sourced through an open wiki, along with bot accounts. The Wikidata data model enables assigning references to every single statement. Currently, there are more than 1 billion statements in Wikidata, of which about 70% have got references. Due to the rapid growth of Wikidata, the quality of Wikidata references is not well covered in the literature. To cover the gap, we suggest using automated tools to verify and improve the quality of Wikidata references. For verifying reference quality, we develop a comprehensive referencing assessment framework based on Data Quality dimensions and criteria. Then, we implement the framework as automated reusable scripts. To improve reference quality, we use Relation Extraction methods to establish a reference-suggesting framework for Wikidata. During the research, we managed to develop a subsetting approach to create a comparison platform and handle the big size of Wikidata. We also investigated reference statistics in 6 Wikidata topical subsets. The results of the latter investigation indicate the need for a wider assessment framework, which we aim to address in this dissertation.