3 ms·
Can you talk more about how you create the questions? I'm guessing the answers are always article titles, and you look for articles that have a neat leading su
by gojomo 15y ago
Can you talk more about how you create the questions?
I'm guessing the answers are always article titles, and you look for articles that have a neat leading summary which is easy to perform anaphor-substitution on. But, how do you pick the candidate wrong answers, and have you developed any other tricks for handling problematic/ambiguous summaries?
- gleitz 15y agoI'll be writing a series of blog posts about the architecture of the game as well as more info about the API at http://blog.wikimaze.me/ http://blog.wikimaze.me/ At a high level you are mostly right. The incorrect answers are generated using http://www.freebase.com/ http://www.freebase.com/ (now a Google product) Lots of pages have to be thrown out, but after analyzing thousands of articles it was relatively straightforward to know when you had a good page versus a bad one.