Presenting the Hunch “academic” API
Today we are releasing a limited form of the Hunch API. We call it the “academic” version, since it will be primarily interesting to academics working in machine learning.
As Caterina mentioned on her blog the other day, Hunch users have answered millions of questions over the last few weeks (about 4.8 million as of today). This means that we have millions of answers to the “taste profiling” questions in Teach Hunch About You, along with hundreds of thousands of feedbacks on Hunch result pages (e.g. a user saying “Yes” the question “Do you like this TV show?”).
We are using this data within Hunch to improve our core algorithm that ranks results. The Teach Hunch About You data is especially important in taste-heavy topics like books, movies, music etc. We will be releasing a new version of our core results algorithm very soon.
Try searching for a Teach Hunch About You question (e.g. type “North America”). Your search result will be a Teach Hunch About You question, along with the percentage of users who clicked on each response. Next, try clicking on a response – you’ll see the other question-answer pairs that are most highly correlated with the response you clicked on.
Note this is just a small part of our eventual API – we will still be releasing a full API in the near future that will let you do stuff like embed a Hunch topic within your website.
Questions? Please ask away in this forum thread.