Personally I just Yelp would have some sort of collaborative filtering engine. They kind of have a social one, but my friends' opinions are often as bad as a random stranger's.
What people review on Yelp is subjective and Yelp makes it hard to filter out people's biases because of their one dimensional rating system. When people review a restaurant, they review the service, the food, the atmosphere and the overall value. For some people, paying $200 on a meal is outrageous regardless of how epic the food might be. For others, they don't even look at the check when they pay and value doesn't factor into their ratings.
Yelp doesn't ask you to rank the businesses you like on a scale of best to worst, so people rate things based on some internal weighting scale that probably shifts slightly from review to review. I may have a 5-star taco trunk right next to a 5-star review for the French Laundry. They aren't equal.
What people review is also temporal. God help a business if there is say, some construction going on across the street, the patron was in a bad mood and becomes argumentative or the chef is sick and food quality dips for a night. Or maybe the waiter just broke up with his girlfriend. Maybe they run out of the one thing the patron really wants to eat. Any of these things can make someone walk away and feel the business is poor.
Businesses and people also change over time, I have reviews on Yelp for 5 stars from 5 years ago for places that have either gone downhill or simply don't match my current taste.
People also tend to only review things they either really liked/loved or hated which results in most businesses (especially in SF) as being rated 3.5-4 stars making Yelp pretty useless without looking at the underlying reviews.
A collaborative filtering engine could fix a lot of these problems. At the very least I'll match people who have my own biases and normalize people's ratings. Of course it still doesn't fix the temporal issues, but I don't think that's insurmountable.
That's an interesting idea. Is the main motivation of your suggestion to find a way to enable you to better discover things you might like without relying purely on reviews and ratings? In general, I agree that ratings and reviews have inherent biases that are hard to smooth out. There are so many more dimensions to it, such as how much weight you apply to a review written by someone that is trying a certain type of food for the first time.
Do you think that you would see better results if Yelp was able to use a collaborative filtering approach to somehow personalize their results such that they more closely match your tastes and preferences - based on other users' data?
What people review on Yelp is subjective and Yelp makes it hard to filter out people's biases because of their one dimensional rating system. When people review a restaurant, they review the service, the food, the atmosphere and the overall value. For some people, paying $200 on a meal is outrageous regardless of how epic the food might be. For others, they don't even look at the check when they pay and value doesn't factor into their ratings.
Yelp doesn't ask you to rank the businesses you like on a scale of best to worst, so people rate things based on some internal weighting scale that probably shifts slightly from review to review. I may have a 5-star taco trunk right next to a 5-star review for the French Laundry. They aren't equal.
What people review is also temporal. God help a business if there is say, some construction going on across the street, the patron was in a bad mood and becomes argumentative or the chef is sick and food quality dips for a night. Or maybe the waiter just broke up with his girlfriend. Maybe they run out of the one thing the patron really wants to eat. Any of these things can make someone walk away and feel the business is poor.
Businesses and people also change over time, I have reviews on Yelp for 5 stars from 5 years ago for places that have either gone downhill or simply don't match my current taste.
People also tend to only review things they either really liked/loved or hated which results in most businesses (especially in SF) as being rated 3.5-4 stars making Yelp pretty useless without looking at the underlying reviews.
A collaborative filtering engine could fix a lot of these problems. At the very least I'll match people who have my own biases and normalize people's ratings. Of course it still doesn't fix the temporal issues, but I don't think that's insurmountable.