I yelp everywhere i go. do you yelp ****?

Water Cooler
#41

I don’t think Yelp reviews are normally distributed, especially for highly and lowly rated reviews. If reviews were normally distributed,then they’d be 3. Not sure how important that assumption is, since Matt is dealing with the extremes (4+ stars) and not the average Yelp establishment.

The number of reviews also impacts the distribution (Yelp reviews tend to be one sided until a large enough sample size). A 4 Star restaurant in NYC is often more consistent and high quality than a 4.5 restaurant in a smaller city, due to NYC having often thousands of reviews resulting 4 stars. As a result,it is much rarer for establishments to be above 4 stars in NYC than other smallerplaces

#42

also a lot of tourists rate nyc restaurants, they dont have refined palates like all the BSDs here.

i always laugh when a garbage chinese restaurant or sushi joint gets 4 or 5 stars from someone out of state when most folks know that place is turrible.

#43

In large markets, you can choose to only see the “Yelp Elite” reviews (ya know, people like me with the finest of palates). That can be helpful in getting higher quality info

#44

had no idea, thanks rawraw

btw does anyone else sort by Latest/Newest?

#45

Are there benefits to yelp elite. The yelp elite manager for Los Angeles is hot

#46

Yelp throws events and pays for stuff, like parties and meet up at restaurants. Look at the “Events” page to see what the LA chapter is up to. But I mostly do it for the girls - when they see that red badge, it is game over.

#47

yea a friend of mine is elite too. he told me a lotta asian girls go.

#48

In the model I described, we are not making an assumption that Yelp reviews are normally distributed. We start with the assumption that they are equally likely (so that there is 50-50 chance that a restaurant will get a high review >3 stars, or low review <=3 stars). Then after observing a rating of R following N reviews, you do Bayesian updating an come up with a beta distribution that describes the probability that a restaurant will get a high review. The beta distribution is approximated by normal when N parameter is large.

So we’ve made a number of crude assumptions (that reviewers are independent and not influenced by ratings year-to-date, that the number of reviews is already relatively large, no adjustments for city/geographic location, how recent the reviews are, etc.). But we don’t assume anywhere at all that rankings are normally distributed. You gotta start somewhere with a basic model and make any refinements and upgrades after. But at its core, this model should give a very crude guidance about what’s the tradeoff between high ranking vs. low number of reviewers. Sure, you can eyeball that 4.8 rating based on 10 reviews is maybe not as good as 4.5 based on 5,000 reviews. But what about 4.8 based on 100 reviews? 4.7 based on 500 reviews? If I don’t have time like Nerdy to search for keywords like ‘grinding’ and ‘twerking’ it will be tough to objectively pick the best restaurant.

#49

Oh OK I must’ve misread your post. I thought it was one of the assumptions. I think it’s cool, I’m bringing it up at the next Yelp Elite event ha ha

#50

I’m saying that good tasting food is at times not very presentable. I’m also saying that some places with good food may not know how to present. Additionally, I’m saying that if the food is presentable, it is very likely they hire a lackluster photographer who is unable to capture the essence of the food.

#51

Ahh good old central limit theorem. I was big on stats in college. Mobious it’s easy man. Just don’t sleep and you’ll have time

#52

The vast majority of yelp pictures are user submitted, so not sure the third point is important.

Of course there are type 1 and type 2 errors. That doesn’t mean the pictures contain no value.

#53

Oh ok got it. I get it if you’re talking about user submissions. I was thinking of restaurant submissions. But, this is yelp. Disregard. Carry on.