In order to deliver great personalized local recommendations, Foursquare needs to understand not only which places are the best, but also what makes places all over the world different from each other. Whether it's a dive bar with great late night dancing or a molecular gastronomy restaurant with an amazing tasting menu, we want to categorize these places and understand the relationship that Foursquare users have with them. That's why this summer we launched “Tastes," short descriptive tags for venues to help users personalize their experience and find places that suit them. Tastes can be as simple as a favorite dish like “soup dumplings" or a vibe like “good for dates".
To better understand what our taste data looks like, I created the “Foursquare Taste Map." Here we see a visualization of the most popular three thousand English tastes. Each taste is connected with a line to others like it, and they are arranged so that similar tastes are closer together. For more technical folks, this is a spring embedding of the k-nearest neighbor graph of tastes using the cosine similarity metric (plotted in Gephi), where each taste is represented as a high-dimensional vector of venue affinities.
Obviously it's difficult to capture all of the relationships between these tastes on a single page, but you can still see amazing structure emerge like “wine island" on the far right, or various niches of Asian cuisine in the lower left hand corner, or a variety of different hubs emerge around common dishes like “seafood", “chicken", and “pizza." We are so excited to have the opportunity to work with this unique data set to better understand all of the places in the world, and thought you'd enjoy this visualization.