Using tag clouds to quickly discover patterns in linked data sets
Xingjian Zhang, Jeff Heflin · 2011
Abstract. Casual users usually have knowledge gaps that prevent them from using a Knowledge Base (KB) effectively. This problem is exacerbated by KBs for linked data sets because they cover ontologies with diverse domains and the data is often incomplete with regard to the ontologies. We believe providing visual summaries of how instances use ontological terms (classes and properties) is a promising route to reveal patterns in the KB and quickly familiarize users with it. In this paper we propose a novel contextual tag cloud system, that treats the ontological terms as tags and uses the font size of tags to reflect the number of instances related to the tags. As opposed to traditional tag clouds, which have a single view over all the data, our system has a dynamically generated set of tag clouds each of which shows proportional relations to a context specified as a tag set of classes and properties. Furthermore, our tags have a precise semantics enabling inference of tags. We optimize the infrastructure to enable scalable online computation. We give several examples of discoveries made about DBPedia using our system.