How are the coordinates of the "beer pins" determined?
While we do know the style of each beer, but we do not use this info in placing the beers on the map. A beer’s location on the map is reflective of how it tastes in comparison to the other 25,000 beers represented in the app. The more similar any two beers taste, the closer their pins will be on the map. Each beer’s unique taste profile is determined by deploying state-of-the-art data mining techniques on our partner RateBeer.com’s database of millions of crowd-sourced reviews. Our process identifies patterns and relationships hiding in the data that even the most seasoned beer geek might not be aware of, and the results are never tainted by clever marketing efforts. We don’t "choose" the locations of anything within the maps, the data dictates it, and so in some sense, you do!
As a simple example of how we use crowd-sourced reviews to build a map of beer, consider three beers A, B, and C. Assume A and B are IPAs while C is a stout. Even the simplest natural language processing algorithms could probably analyze the reviews of all three beers and conclude that A and B are more similar to each other than the similarity between A and C or B and C. Now consider all such possible beer triplets made up of the tens of thousands of beers in the RateBeer database. BeerMapper places each beer in a two-dimensional plane that maximizes the number of correct triplet similarity relationships observed from the text data. This means that the map is not a fixed object but actually changes over time as more reviews and beers are submitted.
Where do colors on the map come from?
The colors displayed on the map are aninterpretation of the color of the beers that are placed on the map. Every pixel in the space reflects the color of the average of the colors of the beers in that space. This is why the map will change over time – adding new beers and reviews for each beer in the database evolves the placement of the pins and therefore the colors on the map. These methods are also used to create the other layers in the app (Bitterness, Maltiness and Alcohol by Volume).
A beer’s color can be measured according to the Standard Reference Method (SRM). This reference point is not always reported by breweries, but by knowing the data on enough beers on the map we can predict the SRM for other similar beers. This process of estimation is more specifically described as combining the low dimensional structure of the beer space with nonparametric regression techniques, inferring the color of beers for which we have no measurement (the technique is also used to predict other missing beer attributes in the data set).
How many beers are represented in the App?
There are about 25,000 beers supported by the app. The beer count is a function of the number of reviews each beer has on RateBeer.com since the algorithms that infer the structure of the beer operate directly on the reviews. We only show beers that have 50+ full-text reviews to ensure enough data points to accurately predict a beer’s unique characteristics. So the more popular the beer, the more likely it has a sufficient number of reviews to be included in the app.
How are the beer styles plotted on the map?
The placement of the style labels is the result of taking the mean location of all the beers that claim to be a part of a particular style. This means that the map and labels will evolve over time as new reviews and beers are added to the database because the location of each beer is completely data-driven by the content of the reviews.
The reason we don’t show borders around the styles is that there is no clear boundary between styles, especially when the focus is on how beers taste; which is precisely what BeerMapper does. For instance, sometimes a beer might be labeled a stout but it actually tastes more like a porter. That beer would be placed by its closest beer matches; which in this case would be in the “porter space.”
How are the recommendations calculated?
A lot of apps claim to make recommendations, but our recommendation engine is truly state of the art - based in theory, modified by practice, with theoretical guarantees. BeerMapper started as a thought experiment for an adaptive optimal design problem in machine learning theory. It uses an ensemble of powerful yet sensitive algorithms that optimize recommendations based on how much information we have on a user’s preferences and how accurate our predictions are to his/her actual ratings. At a high level, the different algorithms are broken down into two classes: feature-based and collaborative filtering. When very little is known about a user’s preferences, more weight is given to what “the crowd” likes. As a user rates some beers, the algorithms begin to learn their particular preferences.
One algorithm used in the recommendation engine finds “like users” who have rated beers similarly. As a user continues to rate more beers, unique taste characteristics are exhibited that can only be captured by the particular properties of the beer itself (no two people have the exact same preferences). In this regime we use algorithms that extract distinct properties of the beer from the text reviews of each beer and learn a model that makes recommendations based on these features in a nonparametric manner.
This means that regardless of how complicated your preferences are, we’ll eventually learn your specific palate given enough ratings. To decide which class of algorithm to use and which algorithm within each class to use at each time, we rely on standard cross-validation techniques to dynamically weight the ensemble of predictions from the algorithms. Thus, we don’t pick one algorithm, we use a little bit of all of them and they dynamically adjust depending on which is working best for a user.
How many beers do I have to rate to get an accurate recommendation?
The quick (albeit probably annoying) answer is “it depends.” The more ratings you enter, the more accurate the recommendations will be for your personal tastes. There are multiple algorithms at play in our recommendation engine, each of which has its own “sweet spot” in the number of ratings needed for making predictions. We adjust the weight each algorithm is given based on how accurate the predictions are to a user's actual ratings, so the recommendations evolve over time with the user. If we had to choose a magic number, we see the best results once a user gets above 12 ratings. If your preferences are more complex than the average user, don’t be insulted as we’ll learn your specific palate soon enough, just keep rating and we’ll do the rest!
What's the difference between beers displayed after hitting the "Get Recommendations" button versus seeing the My Preferences heat map?
The algorithms that decide your preferences are working in a higher dimensional space that cannot be visualized in just two dimensions. The beers that are plotted when hitting “Get Recommendations” are the result of predictions made in this higher dimensional space. The heat map “My Preferences” is a low dimensional depiction of your preferences in this somewhat truncated space. That is, we’re averaging over several different dimensions to bring you the two-dimensional heat map that can be visualized. This is why it sometimes appears as though the recommended beers do not always perfectly correspond with the high-density regions of the heat map.
Why can't I find a certain beer?
Generally this happens because it doesn’t have at least a minimum number of ratings on RateBeer.com or it has recently achieved this number. Don’t worry, we’ll get it added soon. Just to be sure, feel free to give us a heads up and we’ll keep an eye on it (Contact Us). Also, check the beer’s name and make sure its not there under a slightly modified spelling.
Where can I find my recommended beers in stores?
Great question! We will be working on this "next step" for a later version of BeerMapper.
When will this app be available for Androids and Windows phones?
We are working on it! Let us know which platforms are most important to you so we can help prioritize them correctly in the Contact Us tab.
What does IBU mean?
IBU stands for “International Bitterness Units”; which is a scale that shows the expected bitterness in beer, ranging from 0 to 100+ (the higher IBUs, the more bitter the beer is likely to taste).
What does "maltiness" mean?
This is a custom, heuristic measure that has no official definition but often correlates with the sweetness of beers – sweeter beers often taste more “malty.” The exception to this rule has to do with dark beers. Darker beers are generally more “malty“ but not always. For example, stouts such as coffee stouts are often not at all malty and rather bitter. The loose definition of maltiness combined with exceptions to the rule makes the Malitness Layer all the more useful as it shows you how malty a beer is perceived. Not sure if you like this flavor? Don’t worry, we’ll figure it out for you as we learn your palate!
What do the ratings mean in a beer's detail popup?
The ratings we show are from RateBeer.com and they have their own method to create ratings based on the Bayesian Formula. We’ll share a snipit of how they describe it, but for full and up to date description, its best to get the info right from their website: http://www.ratebeer.com/ratingsqa.asp
“A users rating is created on a combination of scales whose total is 50 points. This is divided to create a five-point scale rating. Each beer then carries a weighted average of that 5-point scale rating. This weighted point average is different than RateBeer's consumer-friendly 100-point scale SCORE. A beer's score is based on its percentile ranking among all beers. Every beer also has a STYLE SCORE which is a beer's rank among its style peers.”
How is "popularity" measured in the By Popularity list option?
Popularity is measured based on how many times the beer has been rated on RateBeer.com.
How do I pull up a pin on the map?
To pull up a pin on the map, tap and hold the desired location. Once the pin and beer name is displayed, you can click on the information icon to see more details about each beer. If you are looking for a particular beer, use the search bar function in the upper right hand corner.
How do I delete a beer list?
To delete a beer list you've created, swipe across the beer list to the left. Then touch the red Delete button.