Before Kevin Jamieson was a graduate student at UW-Madison, he walked into a bar one day, unaware that the experience would drive not only his graduate research, but also the creation of a wonderful piece of technology: the BeerMapper App.
At the bar, Jamieson tasted a beer that he liked so much that he went back the next day to have another. The problem was that he did not remember the name. He described to the bartender what the beer was like, and in five guesses and trials, the bartender was able to pinpoint Jamieson’s exact beer from over forty on tap. This sparked his imagination, and Jamieson took the thought and ran with it. “The reason why that was so remarkable was because he was skipping over ones,” Jamieson says. “That means there was a structure. The question was, what was he doing in his head, was he doing some sort of algorithm to figure out which questions to ask next to find the beer?”
Jamieson came to UW-Madison and met Professor Nowak, who teaches electrical and computer engineering and was featured in the professor profile in the March 2013 issue of the magazine. Together, they set out on an adventure with the hope of discovering the science behind Kevin’s encounter. “We put [our data] into a mathematical model,” Jamieson says. “Then we started analyzing it, came up with an algorithm, and we got a result.” The two have since worked together on a lot of adaptive learning problems. “From an engineering standpoint, this type of research falls under categories of signal processing, machine learning, and information theory. Basically, the mathematics of information,” Nowak says. “What inspires me is if we need to collect some data from people, what is the best way to do it,” Jamieson says. Although Jamieson’s experience inspired the research, it only served as a theoretical example while they went through the actual math and research behind this type of thinking. “We said that if indeed this is the model, how many questions would we need to ask, which questions should they be and could we just pick random questions? Or do we have to some how be adaptive? There is some hard science here besides beer … that is what the theory examined,” Nowak says. Wondering if their algorithms would work, the pair was ready to apply their academic paper to the real world.
After the research had been done, the algorithms and hard science discovered, there seemed to be no concrete applications. But with the help of Nowak and others, Jamieson finally decided to apply this science to its original inspiration. “We decided to make an app that does exactly what the bartender did for me,” Jamieson says. The final result is something that fully applies the science of their academic paper to a real world application.
“There is a space to the beer,” Jamieson says. “Stouts are similar to stouts, IPAs are similar to IPAs, and there is some similarity between all these beers; there is a structure.” The process came in through finding that structure and applying it to the app. Using the website www.ratebeer.com, they extracted the reviews for every beer, taking key adjectives from the reviews to relate beers to each other by what the two call “basic dimensionality reduction.” By comparing and contrasting the reviews of two different beers, they have valued the level of similarity between the two beers. The closer the words used to describe two beers, the more similar they are. From this, they came up with a relationship between all of the beers and created a 2-D map of beers, which is the main interface for the app. It is a space with all of the beers arranged based on darkness, bitterness and other properties of beer. This space is very interactive, and it can do many things for the user. “What you can do now is define two-dimensional functions or heat maps over this space that are defining your preferences,” Jamieson says. “What I can do now is ask you, ‘What do you prefer?’” The user can also ask for a recommendation based on preferences or ask for beers that are similar to a certain beer that he or she enjoys or is interested in.
Other useful features are available through the app as well. The user can also generate a heat map of the 2-D space based on their specific preferences. This shows all of the beers and where they fit in two dimensions. The app can also be used as an exploratory tool. For instance, if you are at an unknown brewery, you can find the brewery on the app, and all of the beers that they have on tap will pop up in the 2-D space, showing you what each individual beer is like compared to the user’s preferences. “If you go in to some bars, they have a huge menu,” Jamieson says. “[The menu] is a book with the 200 beer bottle list, and you have to read each description to figure out what kind of beer it is.” The app provides a more visual menu, a way to see what each beer the brewery has is like and where it fits in the 2D space based on what you are currently looking for. Once you find a beer that looks interesting, you can tap on it to see information, reviews, etc.
Because Kevin Jamieson and Professor Nowak are researchers, they are turning the app over to someone who will likely be finishing it for release. The app is an ingenious idea and will enhance the overall beer drinker’s experience. Look for it to be released soon!