That sounds good in principle, but people might get upset ("why did we put
this into Wikipedia then?")A compromise could be to import from (in this example) both Wikipedia and IMDb, add both as a reference to the same claim if they agree, and separately if not. We can then deal with the contradictions manually. On Tue, Apr 2, 2013 at 5:21 PM, Michael Hale <[email protected]>wrote: > I think that's a fine threshold, but it will probably vary some per the > type of data. Our goal is ultimately that every claim will have a > reference, and then we can sort of pass the burden of accuracy on to the > references. Wikipedia has grown well with that principle. Adding references > will be much easier when we batch import data from sources dedicated to a > particular type of data as opposed to parsing information out of Wikipedia > (the IMDb has more consistency checks than Infobox film). > > ------------------------------ > Date: Tue, 2 Apr 2013 10:39:22 -0400 > From: [email protected] > > To: [email protected] > Subject: Re: [Wikidata-l] Running "Infobox film" import script > > On Tue, Apr 2, 2013 at 12:58 AM, Michael Hale <[email protected]>wrote: > > It will definitely have some errors, but I scanned the results for the > first 100 movies before I started importing them, and I think the value-add > will be much greater than the number of errors. > > > Does Wikidata have a quality goal or error rate threshold? For example, > Freebase has a nominal quality goal of 99% accuracy and this is the metric > that new data loads are judged against (they also want to be in the 95% > confidence interval, which determines how big a sample you need when doing > evaluations). > > I haven't looked at this bot, but a develop/test/deploy cycle measured in > hours seems, on the surface, to be very aggressive. > > Tom > > _______________________________________________ Wikidata-l mailing list > [email protected] > https://lists.wikimedia.org/mailman/listinfo/wikidata-l > > _______________________________________________ > Wikidata-l mailing list > [email protected] > https://lists.wikimedia.org/mailman/listinfo/wikidata-l > >
_______________________________________________ Wikidata-l mailing list [email protected] https://lists.wikimedia.org/mailman/listinfo/wikidata-l
