General-world knowledge bases such as Wikidata, YAGO, or the Google Knowledge Graph are important backbones for applications like entity resolution, structured search or question answering. The correctness of these knowledge bases is generally well understood and at the center of attention in automated construction techniques. About their recall, in contrast, most knowledge is anecdotal: While they usually sufficiently cover popular topics, gaps quickly become apparent when going beyond.
In this talk I will discuss the problem of understanding the recall of general-world knowledge bases. I will revisit limitations of the common open-world assumption and standard KB evaluation schemes. I will then present three approaches towards mapping KB recall: (i) association rule mining, (ii) text extraction of counting quantifiers and other cues towards recall, and (iii) entity-centric comparative notions of KB recall.