The Problem:

We all love beer. Yet, today we are faced with an unprecedented variety of options, leaving us overwhelmed and indecisive: What should we drink next?

My Solution:

In our Hack Reactor thesis project, I built a machine-learning server using PredictionIO as a beer recommendation engine. Broadly speaking, the app is based on two main strategies:

  1. Content Filtering Approach: This uses various characteristics of each beer to identify its nature. For example, we utilize breweryDB to determine the style of a specific beer, its alcohol by volume (ABV), and its International Bittering Units (IBU). We then query the database to find beers with similar styles.

  2. Collaborative Filtering Approach: This relies solely on past user behavior, specifically your beer ratings. It employs matrix factorization techniques using the Alternating Least Squares (ALS) algorithm. We characterize both beers and users by vectors of factors inferred from beer rating patterns. A high correspondence between beer and user factors leads to a recommendation.

One advantage of using matrix factorization is its capability to incorporate additional information. When explicit feedback (i.e., your ratings) is not available, we can infer user preferences using implicit feedback, such as your browsing history and search patterns.

As a result, OnTapp matches you with beers that suit your taste. To get a beer recommendation and try out our demo, please visit: http://ontappapp.com/.