On 9/24/2010 8:49 AM, Robin Ryder wrote:
Hi,
Thanks for the quick answers, and for the useful link.
My previous e-mail was not detailed enough; sorry about that. Let me clarify:
- I don't need to crawl the entire Wikipedia, only (for example) articles in
a category. ~1,000 articles would be a good start, and I definitely won't be going above ~40,000 articles.
- For every article in the data set, I need to follow every interlanguage
link, and get the article creation date (i.e. creation date of [[en:Brad Pitt]], [[fr:Brad Pitt]], [[it:Brad Pitt]], etc). As far as I can tell, this means that I need one query for every language link.
The data are reasonably easy to get through the API. If my queries risk overloading the server, I am obviously happy to go through the toolserver (once my account gets approved!).
The first part is easy to do if accuracy doesn't matter. Precision and recall are often around 50% for categories in Wikipedia, so if you really care about being right you have to construct your own categories, and it helps to have a synoptic view. Often you can get that view from Freebase and Dbpedia but I'm increasingly coming around to infoexing wikipedia directly because, for things I care about, I can do better than Dbpedia... Freebase does add some special value because they do gardening, data cleaning, data mining, hand edits and other things that clean up the mess.
Secondly, it's not hard at all to run, say, 200k requests against the API over the span of a few days. I think you could get your creation dates from the history records.