News for hackers. Stuff that matters.
Re:Optimizing trie-based spelling correction algorithms at C... by deepsun on Tuesday July 18, 2017 @08:10PM (#14801121)
I'm curious, how do you guys estimate "fitness" of the search results you found, i.e. whether it was what user searched for or not.

Just looking whether user later clicked on a result, or continued to search?

Re:Optimizing trie-based spelling correction algorithms at C... by gmossessian on Tuesday July 18, 2017 @08:42PM (#14801250)
That’s definitely one signal, but it can be refined by tracking whether the search led to click-throughs, conversions, a different search, a bounce, etc... The conversion is the most important event, but there are a lot of important features aside from that which we'll use to improve results over time. If you're interested, our behavioral api (https://constructor.io/docs/#behavioral-data) goes through some of this as well. We've done a lot of work to figure out which signals are the most important, and are continuing to develop this feature all the time as we acquire more data.