Let the Coding Begin

I have finished my research proposal. I now know what I am doing. I am going to create a website where people can rate beers and receive recommendations for other beers they may not have discovered on their own. The inspiration for creating a recommender system came from using Netflix. I was one of those crazy users who actually rated every single movie I've ever watched. I also rate every song I listen to in iTunes. I like rating things I consume. I also like receiving recommendations. Netflix has helped me discover many great movies I may not have discovered on my own.

As an enthusiast of microbrews, I know how difficult it can be to discover great beers. Many of my favorite beers have been recommended by friends. A fellow enthusiast and Netflix inspired me to create a recommender system for microbrews. With tens of thousands of beers to choose from, the decision making process can be overwhelming. Recommender systems build a user preference profile and make better recommendations over time. The only issue is that most recommender systems are for items that can be mailed or accessed electronically. In most cases, beer must be purchased locally. Therefore it is important to provide recommendations that are likely to be accessible in the user's location. The problem is that information about the user's exact location and the availability of a beer may not be explicitly available. The vast majority of users don't want to take the time to tell the system exactly where they purchased the beer.

The focus of this project is determining how effectively a recommender system can determine the likelihood a beer is available in the user's geographic location. I'll try to explain roughly how my method works. When a user creates an account, they will list up to five cities in which they usually purchase beer. When a user rates a beer, the beer is tagged with all the user's cities. The theory is that the more tags a beer receives of a particular city, the more likely it is to be available in that city. The system will first used a content-based approach to come up with a list of recommendations. That list will then be ranked by the likelihood that it is available to the user and then presented to the user on a recommendations page of the website.

There are a lot of steps involved with making this work so it will be at least a month before even a buggy version of the site goes live. I will post updates as I go along.