On 20/09/2019 17:53, Denny Vrandečić wrote:
...
I have been working on a comparison of DBpedia, Wikidata, and Freebase
(and since you've read my thesis, you know that's a thing I know a bit
about). Simple evaluation, coverage, correctness, nothing dramatically
fancy. But I am torn about publishing it, because, d'oh, people may
(with good reasons) dismiss it as being biased. And truth be told - the
simple fact that I don't know DBpedia as well as I know Wikidata and
Freebase might indeed have lead to errors, mistakes, and stuff I missed
in the evaluation. But you know what would help?
I would also be very interested in seeing this. I had a closer look at
DBpedia recently for a tutorial and was surprised by how different the
data is in comparison to Wikidata. A methodological comparison would
surely be helpful.
Of course, it has to be fair, taking into account that DBpedia editions
are based on a Wikipedia in one language (hence is always missing
entities that Wikidata has). For example, I recently computed the
difference between the following two:
(1) The set of all pairs of ancestors that one can find by following
(paths of) parent relations on EN DBPedia.
(2) The set of all pairs of ancestors that one can find by following
(paths of) mother/father relations on Wikidata, but visiting only items
that are present in English Wikipedia.
I am not sure if this is fair or not, but I found it an interesting
setup (non-local effects of incompleteness) -- and (2) is a nice
illustration of something you cannot achieve in SPARQL on principled
grounds ;-).
Cheers,
Markus