Skip to main content



Yelp vs. Alfred

With advances in smart phone technology and smart phone applications, it is becoming increasingly easier for people to compare two options and decide which one to pick. One of the most common dilemmas I know that I have are trying to decide where to eat when I want to try somewhere I have no gone before. With the help of websites like Yelp, I can see people’s reviews of the restaurants, but how reliable are those sources? Can I really make a good decision based on reviews of others? Will these reviews truly tell me whether a restaurant is good or not?

In the article, “In a Race to Out-Rave, 5-Star Web Reviews Go for $5,” the author mentions the fact that there are perhaps millions of reviews on Yelp that are written purely to boost up the rating of the restaurant. In other words, someone was hired to give the restaurant a 5-star rating. Any restaurant wants to make themselves look as good as can be, but this gives them a false rating—a rating that is very misleading towards an unknowing reader trying to decide whether or not the restaurant will be good. A new application for the iPhone 4s has come out recently that can perhaps help people choose restaurant without the worry of false reviews. Alfred is an application is referred to as “a Pandora for dining.” Users can train this application based on their own preferences. On top of that, Alfred can now be linked the user’s Facebook. With this edition, the user can now see whether or not their friends have “liked” that restaurant. This then gives a much more honest rating to the restaurant as user knows that they’re friend has not been hired to like that certain restaurant.

Suppose you are trying where to eat in an unfamiliar location are trying to decide whether or not a restaurant is good or bad. You read 8 reviews and are given a mix of high and low signals. If Pr[G|S] > Pr [G], where G stands for good and S stands for a series of high and low signals, you will accept. In other words, the probability that the restaurant will be good given a series of multiple signals is higher than your own private signal. Thus, you will choose to go to the restaurant, as the probability seems to state that the restaurant will be good. However, this also means that the amount of high signals is greater than the amount of low signals. In the case of Yelp, the false reviews would greatly affect the result of Pr[G|S]. Generally, most restaurants tend to have above average reviews due to the fake 5-star reviews, so the actually calculated probability would be much higher than it actually should be. Thus, people often find that they are disappointed by the restaurant. In the case of Alfred, the mixed signals you receive will be based on whether or not your friends like that restaurant. While there is a chance that your taste differs from your friends, at least you will be getting an honest review from them so Pr[G|S] should be the true value.

Perhaps most importantly, the fake reviews can almost certainly assure that information cascades are wrong. Suppose you are going out to eat with a few of your friends and you decide one by one whether or not a certain restaurant is good or bad. Your first friend reads reviews on Yelp and determines from those reviews that the restaurant is good. Your 2nd friend has heard from their other friends that that restaurant is actually not good. This friend can follow their own signal, and thus decides that the restaurant is bad. Your 3rd friend is also able to use their own signal as friend 1 and 2 had opposite signals. This friend also read reviews online on Yelp, so they say the restaurant is good. Your 4th friend also follows their own signal and decides the restaurant is good based on the reviews they read on Yelp. In this case, any friend after your 4th friend will say that the restaurant is good. In other words, the cascade has begun, and all friends after friend 4 will ignore their private signals and follow the majority. Even though friend 2 had heard that the restaurant was bad, the overall decision was still that the restaurant was good. If the reviews on Yelp had been true, the cascade perhaps would have fallen on the opposite result: that the restaurant was bad. However, because the reviews are falsely boosted to make the restaurant seem better, any of your friends that read the review will think that the restaurant is good. After the cascade takes effect, any private signal, including friends that have heard from other sources that the restaurant is bad, will be disregarded.

Alfred does not have that problem. In fact, Alfred may not even induce a cascade effect. If you are eating with a group of friends, Alfred can connect to their Facebooks and determine based on the user’s preloaded preferences and restaurants they have liked in the past. Instead of the user themselves deciding which restaurant to go to, Alfred will in a sense choose for you. It will take all private signals into effect rather than disregarding them after a cascade has begun. It is still possible that a cascade effect will occur because Alfred will give a list of recommended restaurants that you and your friends still must choose from. However, at least the private signals of everyone have been taken into account in the initial selection of restaurants. Connecting back with multiple signals in Alfred, it is also more likely that the cascade will be correct with Alfred, as the reviews are honest and real.

References:
http://gadgetwise.blogs.nytimes.com/2011/10/27/making-restaurant-recommendations-smarter/
http://www.nytimes.com/2011/08/20/technology/finding-fake-reviews-online.html?_r=1&ref=yelp

Comments

Leave a Reply

Blogging Calendar

October 2011
M T W T F S S
 12
3456789
10111213141516
17181920212223
24252627282930
31  

Archives