Researchers at KU Leuven university in Belgium have developed machine learning models that can predict consumer appreciation of a brew from its chemical properties, leading to the production of tastier pints.
First, the models were trained on a data set of 250 Belgian beers that included their detailed chemical compositions as well as their scores from a trained tasting panel. The team also added 180,000 public reviews of the selected beers from RateBeer. The team then told the models to identify the components that were important contributors to customers’ sensory perception and appreciation of the brews.
Based on the AI findings, the team was able to alter the composition of the beers in a way that made them consistently score better on overall appreciation in blind taste tests. They believe that their findings could be used “for the production of healthier, tastier products, including novel and improved beverages with lower alcohol contents.”
