This is a 2D projection of ingredient co-occurrence patterns from 1,857 cocktail recipes. Each dot represents an ingredient, and ingredients that appear with similar sets of partners sit near each other on the map.

Tap or hover any ingredient to highlight its three nearest neighbors and see their similarity scores. The map shows the 60 most-connected ingredients in the collection, colored by category.

The method: compute similarity between ingredients based on their co-occurrence partner profiles using Positive Pointwise Mutual Information (PPMI), then project the high-dimensional similarity space into 2D using classical multidimensional scaling (MDS).

The x and y coordinates are outputs of a dimensionality reduction algorithm. They preserve relative distances between ingredients but have no inherent meaning themselves. What matters is proximity: ingredients close together have similar co-occurrence patterns.

Think of it as a map where the neighborhoods are real but the compass directions are arbitrary. Gin and vermouth cluster because they show up with the same partners across hundreds of recipes, not because they share a chemical property or flavor profile.

Like the connection wheel, this map reflects what bartenders and recipe writers have historically mixed together, not what a flavor pairing algorithm would recommend. Thick connections and tight clusters are records of tradition and repetition.

The data comes from Bar Assistant (657 recipes), IBA Official Cocktails (2 recipes), TheCocktailDB (356 recipes) and Mr. Boston Official Bartender's Guide (842 recipes). Every ingredient keeps a link back to the recipes it appears in, and the full methodology is documented on the about page.