Does an IRI on the Semantic Web allow us to say something about the meaning of
the resource it represents? We use type sets and predicate sets as proxies for
meaning. We take the null hypothesis that the IRIs contain no meaning, and the
type and predicate sets cannot be
predicted from the IRI. We then compress the dataset by predicting either the
type (vertical axis) or the predicate set (horizontal axis). The more bits we save, the better we predicted the meaning.
The number of bits
saved by this compression can be interpreted as a significance values. The dotted
lines show the levels of significance at which we can reject the null hypothesis
that the IRIs contain no meaning.