the predictability of deliciousness

Have you ever looked on the right hand side of your facebook newsfeed? There's a little section that has suggested facebook friends – basically they find people that you might know and recommend them to you so that you can add them as a friend. Did you ever wondered how they do that? Apparently, facebook uses complex mathematical algorithms to find people you may know depending on who your friends are and who's page you visit most often (i'm sure a ton of other stuff factor into it too).

Netflix also uses algorithms based on your streaming behavior to create a list of suggested movie and tv show picks. Same for amazon, same for gmail. All these companies employ algorithms to analyze large sets of user data to their advantage. In the election earlier this month, Nate Silver (NYT) used algorithms to accurately predict the outcome of every state in the presidential election.

So what does this have to do with deliciousness?? Well, as it turns out, a scientist named Lada Adamic from the University of Michigan (and facebook apparently) teamed up with two other scientists at Michigan and Harvard to create an algorithm that analyzes a recipe and will be able to tell you if it will turn out tasty or not. This formula can predict with nearly 80% accuracy how many stars your uploaded recipe will receive on allrecipes.com. It can also recommend ingredient replacements if you find yourself short of nutmeg or croutons.

Adamic started by taking the nearly 50,000 recipes and millions of reviews from the site and stripping out every ingredient, method, and taste profile. She then builds a virtual social network of sorts for the ingredients by mapping out the common occurances for each ingredient with other ingredients and the associated flavor profiles and such. The resulting network and data set allows her algorithm to predict with surprising accuracy how a recipe will fare in a taste test (or at least on the site's ratings scheme). It also allows for ingredient substitutions. That way you don't find yourself halfway through a mac'n cheese recipe before you realize you didn't pick up the bacon bits (mm...bacon) and frantically drive down to the store.

While the practical applications of such an algorithm aren't fully explored yet, I could see this being very useful in an interactive site or something that help people find what to cook after work one day. Think of what this could do when coupled with those 30-minute meals shows? You could create an interactive kitchen walkthrough!! Ohh the possibilities...