thenextweb.com
Yelp provides a service I really need in a way and from a company that I really don't like, so I'm delighted to hear about competition!
There is no such thing as an objectively good restaurant. In my town, there are famous pizza places which people absolutely love and then others who think it's an overrated, pretentious mess.
Nevermind attempting to control for different waiters, chefs, people having a bad day, etc. The problem isn't Yelp-specific.
The problem is entirely human.
(That said, the venue ratings product we announced today is globally consistent: everyone will see the same rating for a given place. Our Explore product, however, does make personalized recommendations.)
Imagine if they could build up a ratings profile of people who rate the same place similarly, and then network that out... for instance:
Person 1 likes A and B, dislikes C, and hasn't been to D Person 2 dislikes A, likes B, and C, hasn't been to D Person 3 likes B, C, and D, and hasn't been to A Person 4 likes A and D, hasn't been to B or C
So, A has 2 likes, 1 dislike , B has 3 likes, C has 2 like, 1 dislike, and D has 1 like.
That's the start of a rating scale.
But what if an algorithm could identify that, say, Person 1 and Person 4 have similar tastes... so it could recommend D to 1, and B to 4. It can also see that 2 and 3 have similarity, and recommend D to 2.
Now, here's where it gets a bit tricky. The algorithmn can tease out that A and C are opposites - maybe one has great food, but with bad atmosphere/service, and the other is the opposite.
Thus with that deduction, it can recommend B, but not C to 4.
1. Sparsity of data. People surprisingly rate much less than you think they would. In fact, negative ratings are way less sparse than positive ratings (This to me is unintuitive because this is not how I would act but it is what it is).
2. Lack of features for similarity computation. Sometimes, the rating matrix is all you have to compute similarities or you have crappy metadata. You may turn out to be lucky and pull down a facebook open graph and have enough coverage to work with, it depends on your model.
3. The problem of high variance due to latent features (which you alluded to in the last part): Your model gets harder to track due to in sufficient information as to why a place is good or bad. Maybe, there is a correlation between seasonal variations and special cuisines, maybe they had a shitty chef that one time Person 4 came there.
I am not saying it is not do-able, I am just saying it is hard and sometimes ML fairy dust is not enough. :)
A not-quite-as-good but easier statement to make: People who like place X also like places Y, Z... etc.
Absolutely. So one part of the solution is to stop putting restaurants on an "objective" 1-5 scale, and averaging every human together.
Instead, cluster restaurants so you can "people who liked the overrated, pretentious mess also liked X..."
Low weights: new users, numerous reviews (spamming), low rated reviews
High weights: older users, high rated reviews
Let me guess, you live in Phoenix?
I treat Yelp reviews like I do any online review, say Amazon for an example. Stars don't mean too much but if a place has a ton of reviews and very low average that is probably a good signal. I find a few places that fit the bill and then read the reviews in detail and sometimes the other reviews by people who have outlying reviews. Then I usually cross check at places like OpenTable, Zagat and Chowhound.
I don't recall being disappointed and have found some very excellent spots. I was in Atlanta last week and hit up Kevin Rathbun Steak (http://www.yelp.com/biz/kevin-rathbun-steak-atlanta) and JCT Kitchen & Bar (http://www.yelp.com/biz/jct-kitchen-and-bar-atlanta). Both were great.
Update: I'd say that with Yelp I probably miss out on some great spots that don't have much of a presence there, but I also don't strike out. False positives are worse to me than false negatives. Especially if I'm only in town for a few days.
If Google maps would integrate this then I wouldn't have to deal with Yelp's UI.
Sometimes there are places that I've known forever to be a hidden gem. Now they've become Yelp 5-star and the place gets great business but wait times are always an hour.
Sometimes I'm in a new city and I just Yelp for the latest and greatest. This can lead me to some of the best food I've had, complete with menu suggestions and tips from other users. This is where Yelp really excels. It gives great businesses the business they deserve.
Where Yelp fails me is when restaurants get hurt by harsh and poorly written reviews. A few one stars will even make people avoid a business. I've been to a ton of 3 star restaurants personally recommended to me by a friend and they've been fantastic. When I read the Yelp reviews people will rate 1 star for entirely subjective reasons, even worse for poor service when the explained situation seems completely one-sided. It's one thing to give a highly rated place a second opinion, saying it's overrated. It's another thing to harm an innocent small business and in a way preventing other people from giving the place a chance.
And even with well-reviewed places, for large cities there are hundreds of great places buried in the 4-star <100 reviews list. How many people really scroll past the 5 or 6th page when viewing Most Reviewed and Highest Rated?
The last straw would have to be the extortionist behavior of their sales team. But, that's an entirely different story.
My current local demonstration of the pointlessness of crowdsourced reviews:
http://www.urbanspoon.com/r/71/1709549/restaurant/Melbourne/...
A cafe that's not open yet, has 9 votes including 2 "I didn't like it" votes.
It's interesting that St Ali's other café seems to have a string of recent poor reviews (many written in a suspiciously similar style). Surely it'd be easy for UrbanSpoon to do a quick "sanity check", identify the 9 voters who've voted on a venue they couldn't possibly have been to to form a valid opinion, then remove all their votes and reviews across the board?
That's what _I'd_ do if my objective was a fair and balanced review site. Of course if my objective was a Yelp-like advertising service using poor reviews as a way to blackmail small businesses into paying protection/advertising, I'd obviously _welcome_ poor reviews and vindictive downvotes for businesses that haven't even opened yet… UrbanSpoon, like Yelp, is clearly signalling their intentions to me, and calibrating my expectations of their usefulness...
I live in NYC, and so when my wife and I want to go out to eat or enjoy some activity, factoring in the time it takes to travel (usually 40+ minutes each way to any place interesting in the city), we truly can't afford to tenuously choose the places we dine because we want to hedge our bets that the place we'll be at will be great. If it's not, we just spent several hours going to/at a subpar place when we know there are actually hundreds of really incredible places in the city that we could have gone to instead given we had the knowledge.
It doesn't help that we're both programmers who love to program in our spare time together so even during non-work periods we're also "working" to some extent, so usually on a daily basis the only free time we have is when we're eating or right before bed.
But I digress, TL;DR - Not everyone has the luxury of time and the social prowess to curate locations themselves.
On the other hand, just tracking where I check in on foursquare will be kind of boring. Office, restaurants near the office (which kind of suck, compared to places even a few blocks away), airports, etc. I hope they have some kind of interesting filtering to solve that.
What I'd really like is something built on my actual purchasing history, deeper into the venue than just presence. Knowing that I always get a double double or 4x4 at innout is a pretty valid endorsement. Knowing that whenever I go to Apple stores, I buy Applecare for the products, also useful. It's useful (blinded, statistically) to other people, and presumably could be useful from a loyalty perspective, or just for personal purchase tracking, to me. (I kind of use my Amazon purchasing history like that, now. i.e. "what printer do I have in the office, so I know what toner to buy, when I'm not at the office to check".)
A payment provider (Amex for me, or maybe Square someday) is probably in the best position to do this, actually.
Hardest I've laughed in about a week. +1
I guess the problem is that objective information is hard to standardize in a reviewing system. Plus, I've actually been wondering whether humans would view objective statistics with more faith than human reviews - no matter how ridiculous they are.
(http://www.yelp.com/user_details?userid=kSfj8tii1tw8bH1OYmb-... is the finest reviewer ever in the Mountain View/mid-Peninsula level. 75% of all value I get out of Yelp comes from that person; the remaining 25% is knowing about hours, address, and phone number better than crappy restaurant websites.)
And there's definitely manual rules for offices, homes, airports, and other categories that receive disproportionate numbers of check-ins.
Please don't be tempted to go down the "extort small businesses" path that Yelp seems to, though.
I heard you guys had to move offices during the hurricane to further up in midtown. Were you rolling this out during that move?
Yeah, our SoHo office had no power for almost all of last week. We managed to find temporary office space in Midtown to work out of.
Most of the work for venue ratings was done before the storm though. We've been testing the product internally for a few weeks now. That said, a lot of people were getting work done even from our displaced office space.
Maybe next Foursquare will launch a Pay with (four)Square competitor. I feel like they're uniquely positioned to get this type of purchasing data/history and do something useful with it. By comparison, Square has 75,000 registered merchants accepting Pay With Square [1]; LevelUp has only 3,000 merchants [2]; Foursquare has 1,000,000 merchants.
[1] http://www.nytimes.com/2012/07/19/technology/personaltech/as... [2] http://bits.blogs.nytimes.com/2012/07/12/levelup-zero-scvngr...
I believe credit card companies almost all get deep info when purchasing plane tickets. There may be other categories. Even just knowing main bill vs. tip (which you could easily determine from preauth vs. final amount, if restaurants preauth for the billed amount only, which I think some card associations or acquirers mandate) works as a first approximation -- if I tip 25+%, I probably liked the product or service.
There's also the "purchasing card" market, where itemized invoices (level 2 or 3) get sent to the issuer and then to the responsible party.
Foursquare still has a long way to go before it can consistently give better results than Yelp, but this is a great start! Hope some startups get inspired by this and realize that local search is not a dead problem to work on!
Think of how long it's been since Yelp added any filtering criteria. Take a look at their user-photo gallery functionality. Old, stagnant, and ripe for a new generation. However, since sites have been repeating these mistakes for 15 years, there's no indication that any of these are solvable problems.
Sometimes I feel like companies like foursquare would be better off working on much, much more basic(and less glamorous) problems than ratings and check-ins stuff.
It all fits together my friend. These things just take some time to get done.
-harryh
PS: Imagine you had access to billions of checkins covering 10s of millions of users and places all over the world. What sort of interesting things could you do with that data? Feel like you have some great ideas? http://foursquare.jobs
But what 4sq has that Yelp doesn't is number of visits (per user) and time-relation of visits between places. Instead of ratings, a metric unique to 4sq would be repeat visits (cleaned in such a way to eliminate superficial checkins) and where people go immediately before and after visiting a given place. I would rather see that kind of behavioral data leveraged more than have one more place where people rate things
FTFA:
> The data is based on a bunch of data that Foursquare collects, which it says result in more accurate and varied results than the typical ‘other site’, where ratings can all meet in the middle.
You basically just asked for exactly what the article infers that they're doing already.
So I'll rephrase and say that the rating, as it seems to be presented in the app, doesn't give enough context to show how it is different from how 1 to 10 ratings are done elsewhere.
There is currently no way to explicitly rate a venue on Foursquare.
I don't see a problem wit yelp, reading only a star rating isn't a good way to make it useful, you have to speed read and sample the reviews. At a glance I can tell a reviewer who's fake, or one of the many that just like to hear the sound of their own voice, or they are just an opinionated asshole. But when you speed read a few comments you can pick up general trends. For example if many reviewers all say a hotel was noisy there might be something in it. Ditto for amazon. The "solution" that some have tried is to rate the rating, then use a pagerank-link algorithm to module the effect any one rater on the rating. In other words, if many people who themselves have been rated at rating highly accurately, then my influence on a rating is higher.
The suggestions I have seen here (mainly collaborative filtering) may be of merit, if that's have foursquare plan to do it then with more of a representative sampling of users and maybe it will work well. Currently though they would need a lot more users.
But personally I find yelp very useful, it just requires more effort than glancing at the star rating.
Yelp needs a competitor.
Foursquare needs to be more useful.
The trouble with these rating systems is that they're usually done by one of two groups: people who wish they were foodies or people who had a bad experience.
It's nice to see Foursquare get into this space, but like another commenter said -- it'd be nice to see them tackle the minor issues like hours of operations and menus (maybe through some sort of reward?).
To compare it to tech, it's the difference between someone who understand technology and has a background in it versus a self-proclaimed 'techno-geek' who slobbers over a best buy flyer.