Open Source Tech NYC 201330 Mar 2013
3.30.13 Open Source Tech NYC 2013
###Tianhui “Michael” Li, Data Scientist @ Foursquare
For every 1 degree increase in temperature, people are 2.1% more likely to buy ice cream.
The problem with popularity on Foursquare: it doesn’t tell you what’s good. Example: the ‘Brooklyn Target’ - many reviews, many of them with negative sentiment: ‘I’m in hell.’
The problem with ratings on Foursquare: normalization. A 7 to you might be OK to you, to me it might be great.
The ‘multi-arm bandit’ problem: sticking with what you like can lead to naiveté, an echo chamber vs. trying a lot of things, most of which user won’t like but occasionally will.
Focused on WTR (walk-through rate), not CTR. People vote with their feet.
What does it take to be a good data scientist? Really good probability skills, being able to reason about the world in a world where events happen randomly.