Seminar: Web semantization via dynamic semantics

Our goal is to extend the semantic web foundations to enable describing the semantization process. Considering RDF triples, one can ask where these triples are from: have they been written by human publishers, extracted (e.g., from structured parts of WikiPedia) by rules edited by humans, or by (inductive) programs trained to extract, e.g., subjects (named entities), properties or property values? A typical example is the automated extraction of item properties on a retail web. We refer to several theses containing practical semantization experiments. To describe the reliability of the obtained RDF data we propose a half-a-way’ extension of dynamic logic: programs (extractors) remain propositional, Kripke states are web pages, and there is a lot of reification describing the training and testing data and the metrics of learning.’