Constructing the mapper involved a number of steps. First, Jamieson collected the reviews for about 10,000 beers (each beer must have at least 50 reviews) on RateBeer.com. He had his software parse each review and weight a series of key words that would (ostensibly) accurately describe the beer. Each set of keywords was turned into a vector that represented the beer as a whole, then the software looked for the nearest neighbors in Euclidian distance for each beer vector. The next bit is a bit tricky to understand, but Jamieson basically had his software plot out each beer onto a two-dimensional space, then looked at how "close" three beers were to each other. Doing so allowed him to delineate where one beer style ends and another begins. He did not follow the style descriptors used by RateBeer (because style guides can be notoriously misleading), but let the data dictate where the lines of demarcation would lie.