Making Restaurant Recommendations

One of the main features provided by Foodio54 is the ability to provide restaurant recommendations derived from restaurant reviews. This is the fundamental feature that sets us apart from an average Yellow Page site. We can give recommendations that are statistically better than just using the average rating of a restaurant. Are they always better? No, of course not, but usually they are.

So how does this happen?

Without even thinking about the math that's required for this, let's just think this out logically. How would you get a recommendation for a restaurant without Foodio54? Well you would probably ask a friend their opinion. Would you ask just any friend? No, you would probably ask one that you know from past experience likes restaurants that you also like. You wouldn't ask a friend who hates Chinese food what the best place was to eat Chicken Chow Mein would you? No, of course not.

Foodio54 has the advantage of knowing what thousands of people have reviewed, so we know who likes Chinese food and who doesn't. We can make a determination about how similar your taste in food is to another user based on how you've rated places in the past. Once we know who is similar to you we can ask them where you might want to eat. That sounds simple enough, but there is actually quite a bit of math that makes this work well.

What can make the recommender system more accurate?

More data! The best way to increase the accuracy of your recommendations is to rate more restaurants. Full reviews are certainly welcome, however really you only need to give a 1-5 rating for the algorithm to work. Additionally as we get more data we will tighten the criteria for you to be considered similar to another user. This is a fine line that we are always tinker with, because if we make it too tight we will lose accuracy by being more likely to make wild predictions based on 1 or 2 people, but on the other hand if it's too loose we will just come up with an average because too many people will be giving their opinion.

Who else does this?

• Amazon Recommendations based on sales and views
• Digg Digg has a brand new recommender system
• Google Google's method by which they order search results.
• NetFlix (also see the NetFlix Prize)
• StumbleUpon The ratings from their toolbar give recommended sites to stumble