How in the world did I come up with a juniper berry gose? Is a Chai Saison actually good? A Hazelnut Belgian Tripel? How does this site work?
Great questions. Here’s a bit of what goes on behind the scenes on this site.
First, What to Brew collects a ton of data. Through our website and iOS app, we ask people if a random beer style and a random fruit/herb/spice/whatever is “good” or “gross”. Currently, we have over 60,000 votes. I take all of this data and find interesting ways of using it.
I’m able to look at each style, and each addition, and find combinations that people think will likely work. Since it’s all random, this finds both the common things like chocolate and stout, as well as more unusual ones, like a blackberry English Porter. I’ve never had one, but it sounds delicious.
I’m also able to do some more meta analysis. I have some algorithms that analyze the data to find interesting info. One example is the Similar Additions feature, where I’m able to find additions that tend to work in similar beers. For instance, ginger is currently listed as close to elderflower, cranberry, peach, juniper berries and blackberry. This can be helpful to give you ideas on combinations or substitutions.
Another algorithm looks at additions, and finds what tags from the BJCP Style Guide correlate positively or negatively with an addition. In the case of ginger, it’s likely to work in beers listed as pacific, wild-fermented, aged, spice or fruit, but unlikely to work with beers tagged as eastern-europe, bottom-fermenting, sour-ale-family, bock-family or lagered.
All of this is coming from analyzing crowd-sourced data- so pitch in, and vote!
Do you have ideas for other ways I could analyze this data? Let me know in the comments.