Hi,
thanks for all of these information!
I wanted to clarify the "quality guarantee" in YAGO. It is true that
we did not evaluate every statement individually. Rather, we proceeded
as follows: For each relation in YAGO, we have taken a random sample,
and evaluated it manually (wrt Wikipedia). From the ratio of correct
statements on the sample, we have used statistic techniques to
estimated the ratio of correct statements for the whole of YAGO. This
ratio will usually be lower than the ratio on the sample. Thus, more
precisely, the guarantee reads: With a probability of 95%, the ratio
of correct statements for "actedInMovie" is in the interval
97.36%+/-2.64%.
I also wanted to ask again on the relationship between Freebase and
Wikidata: Freebase was bootstrapped from the infoboxes of Wikipedia,
but I think its main selling point is that volunteers can add and
correct data. Thus, my understanding is that, both in Wikidata and in
Freebase, volunteers would fill up structured, factual information. Is
that right? My intuition is that Wikidata will have a more principled
approach, because it can build on the Wikipedia/Wikimedia culture.
Other comments are appreciated.
Thanks
Fabian
