Querying Wikimedia Images using Wikidata Facts

Abstract

Despite its importance to the Web, multimedia content is often neglected when building and designing knowledge-bases: though descriptive metadata and links are often provided for images, video, etc., the multimedia content itself is often treated as opaque and is rarely analysed. IMGpedia is an effort to bring together the images of Wikimedia Commons (including visual information), and relevant knowledge-bases such as Wikidata and DBpedia. The result is a knowledge-base that incorporates similarity relations between the images based on visual descriptors, as well as links to the resources of Wikidata and DBpedia that relate to the image. Using the IMGpedia SPARQL endpoint, it is then possible to perform visuo-semantic queries, combining the semantic facts extracted from the external resources and the similarity relations of the images. This paper presents a new web interface to browse and explore the dataset of IMGpedia in a more friendly manner, as well as new visuo-semantic queries that can be answered using 6 million recently added links from IMGpedia to Wikidata. We also discuss future directions we foresee for the IMGpedia project.