Your user knows they want a healthyish but tasty pasta for dinner but aren't quite sure exactly which recipe to choose. How can you help narrow their search and show them closely related recipes to give them enough options without making their search exhausting? This talk will show you BuzzFeed/Tasty tech's solution to creating a consistent method for finding similar Tasty recipes using word2vec.
The Tasty Tech team at BuzzFeed launched an app and website this year. We knew that to organize our 2+ year backlog of recipes into compelling collections for our app and site that we'd need to know more about the content of those recipes. We experimented with having producers tag and categorize our recipe and compilation content and (not surprisingly) humans had a lot of varied ideas about what constitutes healthy versus comfort food and what's an "easy" recipe. To overcome this, we developed a machine-readable method for describing a recipe using Google's word2vec, a tool for computing vector representations of words. We used these word vectors otherwise known as word embeddings to create vector representations of each of our recipes that encoded information about the preparation and ingredients of each recipe. These vector representations, dubbed recipe2vec, feed into our related recipes API which determines how related a pair of recipes is using cosine similarity. This powers our feed at the bottom of each recipe on the site that surfaces related recipes to the main recipe and will help power many data and machine learning products to come at Tasty.