Hacker News Reader

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?

View on HN

I'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/

At a high level you are mostly right. The incorrect answers are generated using 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.

Reply on news.ycombinator.com