Description: This passage is mainly about okcupid login. This passage is from a talk given at the American Sociological Association’s annual meeting. In this passage, the writer talks about modern romance which is called OK Cupid.
This is one of the founders of OK Cupid and we’re talking about modern romance tonight, and I know that no discussion of that is complete without talking about online dating, so I think I will show you that in my ten minutes, I look under the hood at how a dating site recommends people to one another.
It’s because they are all dating sites, I like a big marketplace or a big warehouse for the people, and you get to figure out that 10 New Yorkers of that 1/2 million you have to show when someone refreshes the page, this is how we do it, I’m going to figure this out, I’m going to do that, this is showing up there, it is perfect.
So let’s talk about the first component, now for the first one that we think of is called master percentage, this is a simple Shannon answer system, it’s what you expect when you go to a dating site especially back, we start it, keep it in 2003, you can see some questions about if you want to have kids, if you believe in God or if you smoke or have a foot fetish.
You can answer whether you do it, you say how you want other person to answer and then we do a little bit of basic arithmetic, they get some scores for the two of you and I click agreement, you don’t have to agree, you don’t have to both answer in the same way, you can stipulate a asymmetric type of answer.
So a lot of people who are sexually dominant don’t want another dominant person, that’s going to be a lot of awkward collisions, so you want another submissive person, and we also let you know that this is irrelevant and you can skip the question if you don’t care about dogs or having kids, so this is the foundational idea of OK Cupid.
This is the thing that we put on a proverbial nap on the proverbial napkin, that’s going to be a sweet dating site, it turns out that, this is not sufficient to run a dating site, everybody has some versions of this thing, they spend all their time on this, I’m assuming the other two cases and I know about matching dinners intimately.
Because they’re also owned by the same company that owns OK Cupid, to be clear I’m not speculating you, you start your dating site or dating out and you discover quickly that no matter what you’re doing or what people are typing them about themselves, everyone swipes left or bypasses the people and clicks on the hot people, it’s what we call it Ok Cupid.
It is the focus problem and it’s a problem because you end up with a situation where one person who always is a woman in the heterosexual case gets too much attention, she can’t deal with it, she can’t reply to the messages, she’s grossed out, she maybe leaves or ignores everyone, so at that point they have a choice, they can either get bummed and leave two or DoubleDown.
I’m going to send more messages and then they do it to the next person, they’re the whole thing such as death spirals, it’s a big bad deal, I don’t know if I’m turning on my slides, so in order to circumvent this, you need to know who is good-looking before they run into this problem, so you need a way of capturing attractiveness outside of the recommendation engine itself.
For us it’s called quick match, this is the earliest version of it, I can find that it’s a little too blurry to read, this is from 2005 or 2006, I think you can’t read this purple over there, but we have wiki edits for people’s profiles, we have like points that you can earn, we have tried every single thing under the Sun including this which works these days.
We have realized that through a series of experiments it’s only the picture that matters, so all this other schwag such as their profile has gone, you let people rate looks and personality separately, I am also naive, they are almost the same for every single person on a per-capita basis, so we get rid of that.
I am going through the old logs when I am running my book, here’s someone with a four and a half star personality, so it’s the 95th percentile, I look at the account, it is a woman in a bikini hugging this piece of wood with no text and everyone’s like that, she’s so awesome to talk to, so they cut that up.
So this the rating information, this is the most important single piece of data that we use, it’s how we make a lot of decisions and the next thing is to attract in this distance, so when we’re recommending you want that distance between the two of you to be as small as possible, and that’s how it’s weighted, and this speaks for itself, I won’t read the PowerPoint to you.
But as an implication, it’s a next-level type of advice, if anybody ever complaints you that anyone everyone on tinder OK Cupid, it’s ugly, it means that they are ugly also, because you have to run the site after many years of experimenting and seeing what drives, and most conversations are homogeny.
I have learned from Eli over a club soda earlier, it is very important and here it is to run through this data quickly, our table now is four billion rows of votes, Tinder’s is probably ten billion so this is man rating women, it’s well centered such as bell curve, the natural median should be three, the observed meaning here is two point eight six which reflects our prior belief of how attractiveness is distributed genetically out there.
You have a lot of ugly and mediocre people, you have some very hot people and some veryugly people, and you know that’s how it should work, in the most dating sites the women are rated that the same weight in the center tinder matched the same on OK Cupid, this is how the men are aided, so it is substantially lower.
So there are two things that can be working here, there’s obviously a market dynamic, there are slightly more men than women, but I think it’s also the timbre of the site, I have this thesis on match.com, these are the blue and the red curves which overlap the proposition, let’s go to have some casual sex or you want to come over to my spooky apartment.
Let’s go on a date, I’ll give you a rose or the women and the men rate each other about the same on tinder which is an extreme way, that blue curve is jacked way on the y-axis such as superhighway which tops out 40 percent, I have less than five percent yeses, I’m plowing through, I know we’re short on time.
This data is also something that academics asked from us a lot and you can do fun things, because we collect so much demographics, such as race, age, sexual, orientation and body type, I’m going to run through age quickly, this thing works so this is how women rate men by age, I know it’s hard to read, it is probably small.
So you see women are 20, 25, 30 and 35, the block is the highest decile to the women of that age, so 25-year-old women rate 26-year-old men to be the hottest, 30-year-old women rate thirty-year-old men to be the hottest, I think you can see that it’s a diagonal, so women think that their own ages are better-looking until they are 39 or 40.
That’s good, don’t get older, this is how German records, that speaks for itself, you can play with this stuff forever, and that’s how I end up writing my book, there the third vector is responsiveness, so we take this, the third factor that we have added is that we prioritize users with fewer unread messages.
So even if you’re the same level of hotness, even if in theory you like the same number of things or your match percentage is good, we want to send you to people that are going to write you back message, health is extremely important, and by health everybody gets messages in depth, everybody gets a lot of messages, we don’t want it all piled in.
It’s another focus problem, it speaks for itself and then randomness, we add a big random number and it sounds a little counterintuitive, we would essentially mathematically undo the other three variables by adding a big spoof into our formula, the rationale is as follows, one there should be some serendipity, and that’s madness, that’s the mathematical serendipity.
You can’t get that, first of all we don’t think that we have calculated compatibility exactly, if you think that when someone hits OK, OK Cupid matches me, you can’t show them the same ten people because they’re the the ten best, they’re going to think that nobody uses your site, so you have to keep stirring the pond.
You have to intentionally show people that might not be the ice-cold number one match for somebody according to the other three criteria, because if you do that, people will leave your site because it will be so still, so that’s how we do it, this is how we weight them, this is by far a path attractiveness similarity.
The biggest of the other three things that we think are that we re geniuses for thinking of stupidly enough number two when we think of OK Cupid, which is much smaller and all of this comes after a filter, so we have to spend a lot of times, you have to obey what the user tells you, even if it might not be in his or her best interest.
But if somebody tells us that they only want to see twenty five-year-old people within five miles of me, you get a filter that craps out, you do a hard filter and then you sort the results and you spit them back, a bunch of profiles will sleep with it, so we’re trying to get people to talk to one another.
There’s never been a discussion of finding you a soulmate, I don’t think anybody in the industry thinks about debunking anyone who might have that idea, and I think the ultimate algorithm obviously is always going to be in each individual person, there is no website such as somebody’s funky breath or a cool sweater or awesome haircut or any of that stuff that like make or breaks that, first make or break a first date, so thanks I’ll talk more later.