"Using external knowledge graphs to enrich texts and images with background knowledge "

The vast majority of knowledge that is produced daily within our society is in textual format. But at the same time, efforts are being made to publish structured, factual knowledge, such that today there are thousands of public, structured data sets, containing billions of facts. These facts correspond to background relations between concepts that texts very often do not make explicit. Relevant facts extracted from structured data sets can therefore enrich documents so that on the one side, they support humans in interpretation tasks, and on the other side, they make it easier for machines to relate, organise, retrieve, and use such texts.
In this talk, I will present our work on exploiting external knowledge graphs for quantifying and explaining relations between concepts. These relations can then be used for disambiguating entities and common nouns. Furthermore, I will show that relevant hidden knowledge can be revealed even when starting with as little data as an image and its caption. In this direction, I will present how we make use of the knowledge graph in order to identify the intended gist behind images.
The particularity of the methods I will discuss is that besides achieving high accuracy, they also provide explanations for their results in the form of knowledge graph excerpts.