This paper investigates to what extent beer-related data on the ratings platform Untappd can inform us about which factors have the largest influence on consumer ratings and how tastes have changed over time. A random forest model was trained to predict the chance of a highly rated beer with 88.1% accuracy confirming the two most important predictors of rating were a high ABV and specific beer styles (i.e., Lambic). A Latent Dirichlet Allocation algorithm was also trained to identify topics of tasting notes, revealing that popularity for fruity and tropical tasting notes has grown rapidly since 2010 at the expense of more bitter beers.
For more details please see the pdf report above.
Data source: https://untappd.com/