|
Are you in any way involved with the blog? Because I just wanted to say that it's really one of the most interesting reads I've seen. Everyone else: proclick
|
| # ? Apr 13, 2011 02:01 |
|
|
|
| # ? May 19, 2013 18:50 |
|
|
Do you know if they share data with researchers? I study online privacy and am very interested in (anonymized) data sets. We have people who develop crazy statistical techniques to piece together innocuous information to identify people, we'd be able to make OKtrends look like some crappy Xanga.
GregNorc fucked around with this message at Apr 13, 2011 around 02:58 |
| # ? Apr 13, 2011 02:20 |
|
|
This is interesting. Thank you for posting here. Question: Roughly what's the algorithm for quiver matches? Is there anything to it more than "three random people that meet your criteria and match above ~80%"? Suggestion: In one of the oktrends, they used data from that box where you can type the profile name of a person you met on okc as your reason for deactivating a profile. But the thing is, your most loyal-or-whatever users will probably just set their profile as "seeing someone" and go inactive rather than actually disabling it. There's more data to be had if you were to add a box to type a name in when you select "seeing someone", with an option to display this publicly or not. I realize this isn't your area, but this falls under "is it ever going to happen". Suggestion: Also not really your area, but based on my time in the other okc thread, some of your users would probably be helped by more options for these profile fields: sex/gender/looking-for/interested-in: people of various sexual persuasions not quite hetero/homo/bi seem pretty torn on how to handle these. To be fully progressive, they might even be better served with a write-in box for people to use whatever sexual label they please. Heck, even having a field for this rather than just taking it from your match preferences would probably help. drugs: something to differentiate between "smoke weed everyday" and "do heroin everyday" since right now both want to say "frequently." The standard seems to be that people who do weed only write "---" but such an unspoken convention is hardly user friendly. [edit] Oh, yeah, that one below really messes with users a lot. I "vote" for it too. Lakitu7 fucked around with this message at Apr 13, 2011 around 06:43 |
| # ? Apr 13, 2011 04:29 |
|
|
How would I, an internet user living on a border of two countries, restrict my matches to only one of those countries.
|
| # ? Apr 13, 2011 06:32 |
|
|
What's the deal with the occasional math or literature match question that pops up that I can't skip out of? Are these anti-spam measures, or are you just really interested in the average IQ of an OKCupid user?
|
| # ? Apr 13, 2011 10:52 |
|
|
ChrizpG posted:I guess you probably don't talk to engineering about this but I'm curious as to what sorts of machine learning techniques are applied to that glorious data. I figure you can't be too specific or whatever but it'd be neat to know if its a sophisticated neural network that knows who I'd like to date. I have no idea, I'm sorry! The best I can give you is a slightly nerdy breakdown of how the match percentage algorithm works: http://www.okcupid.com/help/match-percentages
|
| # ? Apr 13, 2011 12:19 |
|
|
quote:Are you in any way involved with the blog? Because I just wanted to say that it's really one of the most interesting reads I've seen. Everyone else: proclick nope, that's my buddy Christian. He's awesome! quote:Do you know if they share data with researchers? I study online privacy and am very interested in (anonymized) data sets. We have people who develop crazy statistical techniques to piece together innocuous information to identify people, we'd be able to make OKtrends look like some crappy Xanga. Nope, they absolutely don't share the information with anyone, even anonymized. quote:This is interesting. Thank you for posting here. you're welcome! quote:Question: Roughly what's the algorithm for quiver matches? Is there anything to it more than "three random people that meet your criteria and match above ~80%"? any suggested match on the site is simply match % and what you enter on your profile under "Looking For". The algorithm will tweak itself if one person rates another highly, if one person adds the other to their favorites, if the person is a harder-to-match age or location on the site, etc. This goes for quiver, quickmatch, "you might like", etc. quote:Suggestion: In one of the oktrends, they used data from that box where you can type the profile name of a person you met on okc as your reason for deactivating a profile. But the thing is, your most loyal-or-whatever users will probably just set their profile as "seeing someone" and go inactive rather than actually disabling it. There's more data to be had if you were to add a box to type a name in when you select "seeing someone", with an option to display this publicly or not. I realize this isn't your area, but this falls under "is it ever going to happen". Hm, that's a good idea and I'll pass it along. We don't really use that data publicly, so I'm not sure if it's the most interesting data set that they'd want to create, but it might be! quote:Suggestion: Also not really your area, but based on my time in the other okc thread, some of your users would probably be helped by more options for these profile fields: I have absolutely already brought adding queer/trans/etc up to my bosses and tried to push for it as much as I can, but I'm just a lowly work-from-home peon. Write-in-your-own would mess with the database and make it impossible to search for any particular term (you'd have like, guess what people may have put instead of using a dropdown menu) but definitely adding some more options makes sense. quote:drugs: something to differentiate between "smoke weed everyday" and "do heroin everyday" since right now both want to say "frequently." The standard seems to be that people who do weed only write "---" but such an unspoken convention is hardly user friendly. Yeah drugs is a hard one- people have suggested having weed as a separate 'non-drug' category, but I'm not sure if it's going to happen. quote:How would I, an internet user living on a border of two countries, restrict my matches to only one of those countries. quote:What's the deal with the occasional math or literature match question that pops up that I can't skip out of? Are these anti-spam measures, or are you just really interested in the average IQ of an OKCupid user? Yep, those are staff questions to see how smart people really are. You'll notice one of the questions is "do you think you're more intelligent than the average user" and then you get a bunch of math/ SAT type quetions. The best is "Is the earth bigger than the sun?" haha
|
| # ? Apr 13, 2011 12:34 |
|
|
I don't want to be that guy that laments things were "better back then", but has it been a conscious decision to move the site from its more tongue-in-cheek caustic approach to a more "friendly" version? Or has it just been a natural evolution? "Stalkers" -> "Visitors", the "less independent" award became "more attentive", and the rarest unicorn of them all, the "More/Less Radcliffy" award has entirely vanished! (On that note, I really want to know what questions lead towards being more/less Radcliffy. I tried and tried and tried to figure it out, but never got anywhere.)
|
| # ? Apr 13, 2011 13:05 |
|
|
devilmouse posted:I don't want to be that guy that laments things were "better back then", but has it been a conscious decision to move the site from its more tongue-in-cheek caustic approach to a more "friendly" version? Or has it just been a natural evolution? I have no idea. I've just been working for them for less than a year so I can't really speak to how things used to be. I can say that becoming more user-friendly is something that happens to most companies as they get bigger! Especially now we have a LOT more 50+ users than even a year ago, and ALL of them get bent out of shape about the stuff you mentioned above. It seems like the path of least resistance to just smooth things out a bit.
|
| # ? Apr 13, 2011 13:38 |
|
|
aliceamadee posted:nope, that's my buddy Christian. He's awesome! edit: v Generally speaking, the "correlation proves causation" fallacy doesn't bother me so much. In a lot of cases, correlation is sufficient for the analysis to be useful. One of my favorites comes from this post: http://blog.okcupid.com/index.php/t...or-first-dates/ where asking a girl "Do you like the taste of beer?" has a strong correlation with whether she will have first date sex. Causation is not necessary, and in fact is not even germane. Josh Lyman fucked around with this message at Apr 13, 2011 around 14:51 |
| # ? Apr 13, 2011 14:37 |
|
|
Josh Lyman posted:Do you know what Christian's background is/whether he works full-time? I absolutely love the OkTrends blog, but I'll periodically have qualms with his data analysis techniques (I do very similar work as a PhD student). You mean the fact that it is one giant data mining correlation = causation masturbation?
|
| # ? Apr 13, 2011 14:40 |
|
|
Christian has a wikipedia page if you want to read about him: http://en.wikipedia.org/wiki/Christian_Rudder
|
| # ? Apr 13, 2011 14:56 |
|
|
aliceamadee posted:Responses aliceamadee posted:There's no way to completely restrict your matches to one country (or even one city). We know it's an issue, but I'm not sure if there's a fix on the tables right now. Sorry! aliceamadee posted:Yep, those are staff questions to see how smart people really are. You'll notice one of the questions is "do you think you're more intelligent than the average user" and then you get a bunch of math/ SAT type quetions. The best is "Is the earth bigger than the sun?" haha Josh Lyman posted:Do you know what Christian's background is/whether he works full-time? I absolutely love the OkTrends blog, but I'll periodically have qualms with his data analysis techniques (I do very similar work as a PhD student). Chin Strap posted:You mean the fact that it is one giant data mining correlation = causation masturbation? I love oktrends and recommend it to everyone, but when I link it to people with in a more statistician-type field I have to preface it with "Well I know some of the stats are iffy but it's still a great read...". While it wouldn't take care of everything, it would be a lot of improvement for very little effort if he would just add confidence intervals to his graphs / statements. While there'd still be an element of correlation/causation, this is common to any such data-mining literature, and at least we'd be able to see the relative strengths of said correlations.
|
| # ? Apr 13, 2011 17:56 |
|
|
Lakitu7 posted:Thank you for your replies! I know most of it isn't your department, but it's nice to be heard. Admittedly, I can also see how adding 20 more options for sexual orientation and acknowledging marijuana would be a bit confusing to the growing older audience you discussed too. As a user I'm like "go back to eharmony you prudes" but that's not a good business idea for you guys obviously. yeah, I mean OkC isn't really intentionally trying to become the new eharmony or anything, and personally I am all for adding lots of options for things, but you'd also be surprised at the number of emails I get from FREAKED OUT DUDES just because another dude LOOKED AT HIS PROFILE ONCE. Lakitu7 posted:distance of users Honestly I don't know what the programmer's reasons are for only having >25 miles but I assume they have to have one because we've definitely gotten a lot of comments about wanting to limit searches by a city/ country/ <25 miles. Lakitu7 posted:So, wait, are you saying that how you answer some questions directly affects what sort of questions it asks you in the future? no no, there are just some questions that everyone gets asked and can't skip, where others are randomly generated and are skipable.
|
| # ? Apr 13, 2011 18:09 |
|
|
aliceamadee posted:yeah, I mean OkC isn't really intentionally trying to become the new eharmony or anything, and personally I am all for adding lots of options for things, but you'd also be surprised at the number of emails I get from FREAKED OUT DUDES just because another dude LOOKED AT HIS PROFILE ONCE.  " It'd be hilarious if you could share some of those but you probably can't.  aliceamadee posted:Honestly I don't know what the programmer's reasons are for only having >25 miles but I assume they have to have one because we've definitely gotten a lot of comments about wanting to limit searches by a city/ country/ <25 miles. aliceamadee posted:no no, there are just some questions that everyone gets asked and can't skip, where others are randomly generated and are skipable.
|
| # ? Apr 13, 2011 18:26 |
|
|
Lakitu7 posted:Haha nice. "I AINT NO QUEER TELL ALL THE QUEERS NOT TO LOOK AT MY PROFILE FOR ME hah yes, exactly! Some people get OUTRAGED that they can't block all men from seeing their profile! We did actually recently institute the opposite of that, which is that gay people can block all straight people from seeing their profile (especially good for gay women who get harassed by 23082308 dudes trying to get them to have a threesome). It's been suggested that we have a "homophobe" setting which will block everyone who's not straight and of the opposite gender (it's even been suggested that it's automatically applied to anyone who thinks gay marriage is a bad idea and answers so in a question), but I don't know if it actually happen or not. I think it would be pretty sweet.
|
| # ? Apr 13, 2011 18:34 |
|
|
edit; rant not worth potential derail
Dross fucked around with this message at Apr 13, 2011 around 18:55 |
| # ? Apr 13, 2011 18:52 |
|
|
I'd say in some cases it has nothing to do with homophobia, it's just that they simply don't want to be seen by their male peers on the site.
|
| # ? Apr 13, 2011 18:55 |
|
|
Rick Rickshaw posted:I'd say in some cases it has nothing to do with homophobia, it's just that they simply don't want to be seen by their male peers on the site. The basic principle of OkCupid is that if you want to be able to view/ message anyone for free, they should be able to view/ message you as well. If you're in a bar looking to pick up chicks, you can't stop dudes from seeing you.
|
| # ? Apr 13, 2011 19:03 |
|
|
aliceamadee posted:The basic principle of OkCupid is that if you want to be able to view/ message anyone for free, they should be able to view/ message you as well. If you're in a bar looking to pick up chicks, you can't stop dudes from seeing you. While true, there's not really a stigma towards going to a bar and picking up chicks. Unfortunately for online dating, there still is. It is getting eaten away at year by year, though. But I'm not suggesting OkCupid go through the trouble of making it a feature to hide your profile from dudes.
|
| # ? Apr 13, 2011 19:09 |
|
|
That and being able to scan other dude's profiles for sweet lines is a godsend for some people.
|
| # ? Apr 13, 2011 19:09 |
|
|
I ditched OkCupid because of the insane amount of bisexual girls listing themselves as gay to "hide from men". Does this feature now extend to bisexuals so they don't lie anymore? Thanks for starting this thread! I think the site is really cool even if my experience wasn't.
|
| # ? Apr 13, 2011 19:43 |
|
|
Ohh man, I am totally going to set my orientation to bi and start viewing a bunch of straight dude profiles and see how much rage I generate.
|
| # ? Apr 13, 2011 20:06 |
|
God I fucking love Diablo 3 gold, it even paid for this shitty title
|
I know you go after spammers, but what is the policy for people who have pretty obviously fictional profiles? I had an OKCupid profile years ago that I wrote from the perspective a Bond type superspy (I think I listed dangling one handed from a helicopter as a hobby). It was very tongue in cheek and my most private thing was that the profile "might not be 100% accurate". I never got banned. Was that an oversight or is this a valid profile option?
|
| # ? Apr 13, 2011 20:29 |
|
|
Demon Allie posted:I ditched OkCupid because of the insane amount of bisexual girls listing themselves as gay to "hide from men". Does this feature now extend to bisexuals so they don't lie anymore? yes, bisexual users can hide themselves from straight users if they wish. Grifter posted:I know you go after spammers, but what is the policy for people who have pretty obviously fictional profiles? I had an OKCupid profile years ago that I wrote from the perspective a Bond type superspy (I think I listed dangling one handed from a helicopter as a hobby). It was very tongue in cheek and my most private thing was that the profile "might not be 100% accurate". I never got banned. Was that an oversight or is this a valid profile option? fake profiles/ more than one profile is against the Terms of Service. However, if it doesn't set off any of our triggers and no one flags it, you'll fly under the radar. If I see an obviously fake profile though, I delete it.
|
| # ? Apr 13, 2011 20:57 |
|
|
What's the deal with the 'top half of attractive people' email? The general feel seems to be that it's randomly sent out (not that goons are an unbiased sample pool).
|
| # ? Apr 14, 2011 07:04 |
|
|
aliceamadee posted:hah yes, exactly! Some people get OUTRAGED that they can't block all men from seeing their profile! We did actually recently institute the opposite of that, which is that gay people can block all straight people from seeing their profile (especially good for gay women who get harassed by 23082308 dudes trying to get them to have a threesome). It's been suggested that we have a "homophobe" setting which will block everyone who's not straight and of the opposite gender (it's even been suggested that it's automatically applied to anyone who thinks gay marriage is a bad idea and answers so in a question), but I don't know if it actually happen or not. I think it would be pretty sweet. Have you read this article: http://www.theatlanticwire.com/nati...ou-dates/18034/ specifically referring to this bit: quote:I asked Yagan whether OkCupid might try tailoring its algorithm to surface more statistically successful racial combinations. Such a measure wasn’t out of the question, he said. “Imagine we did a lot of research, and we found that there were certain demographic or psychographic attributes that were predictors of three-ways. Hispanic men and Indian women, say,” Yagan suggested. “If we thought that drove success, we could tweak it so those matches showed up more often. Not because of a social mission, but because if it’s working, there needs to be more of it.” And if so, how do you feel about it, especially in light of the 'homophobe' setting suggested above? minya fucked around with this message at Apr 14, 2011 around 07:12 |
| # ? Apr 14, 2011 07:10 |
|
|
Chin Strap posted:You mean the fact that it is one giant data mining correlation = causation masturbation? Well that's an appropriate avatar. As far as I can recall, every single post is a collection of descriptive statistics and graphical techniques - there isn't a single claim of causation. Hell, there isn't even a single t-test (though I would be stunned if much of it isn't significant with n often surpassing 1,000,000,000). I think his use of statistics is actually quite suitable for his goals: to show something interesting, entertaining, and most importantly, accessible to a lay public.
|
| # ? Apr 14, 2011 12:14 |
|
|
Mausi posted:What's the deal with the 'top half of attractive people' email? The general feel seems to be that it's randomly sent out (not that goons are an unbiased sample pool). It has to do with your star ratings. It's definitely real but any user can still find any other user- you just have to use Match Search instead of relying on homepage matches/ quiver/ quickmatch. minya posted:Have you read this article: http://www.theatlanticwire.com/nati...ou-dates/18034/ I think it's interesting. However, I think there's a big difference between being able to block people of a gender you'll never be sexually interested in, and showing/ hiding people of certain ethnicities. There are definitely racists on OkC-- just like I get messages saying "WHY DID A DUDE LOOK AT ME" I also get messages from people saying "WHY AREN'T YOU ONLY SHOWING ME WHITE PEOPLE" or "WHY ARE YOU SHOWING ME SO MANY FAT PEOPLE". I think that's a big reason why the quiver/ quickmatch/ homepage doesn't take physical characteristics into account at all. If people want to search by something crazy specific, they can, otherwise it's left up to personality and general compatibility.
|
| # ? Apr 14, 2011 12:26 |
|
|
Forever_Peace posted:Well that's an appropriate avatar. As far as I can recall, every single post is a collection of descriptive statistics and graphical techniques - there isn't a single claim of causation. Hell, there isn't even a single t-test (though I would be stunned if much of it isn't significant with n often surpassing 1,000,000,000). I think his use of statistics is actually quite suitable for his goals: to show something interesting, entertaining, and most importantly, accessible to a lay public. There isn't a single measure of standard error on anything he posts. "This number is bigger than this other number" doesn't help you when you don't know error bars on those numbers. And how much is looked at that is subsequently thrown away because it doesn't look like anything? I could make 1000 random plots of noise, find 5 that look like they have a sort of signal, and make a blog post about it. Doesn't mean that it is interesting or has any sort of statistical rigor.
|
| # ? Apr 14, 2011 13:03 |
|
|
Chin Strap posted:There isn't a single measure of standard error on anything he posts. "This number is bigger than this other number" doesn't help you when you don't know error bars on those numbers. And how much is looked at that is subsequently thrown away because it doesn't look like anything? I could make 1000 random plots of noise, find 5 that look like they have a sort of signal, and make a blog post about it. Doesn't mean that it is interesting or has any sort of statistical rigor. I think it's because if they included all that, it wouldn't be a fun article so much as it would be something out of a scientific journal, which nobody likes to read. The guy doesn't claim perfect causation, just correlation, and he presents it all in a way that the common person can understand AND find interesting. "Why doesn't 'Go Dogs, Go!' have data and analysis explaining the evolutionary mechanisms by which dogs had gained thumbs and the ability to drive???" 
Fnoigy fucked around with this message at Apr 14, 2011 around 13:22 |
| # ? Apr 14, 2011 13:20 |
|
|
How does that system take into account the fact that many people don't use the full spectrum of ratings? For example, I don't think I've ever rated someone two stars. I use the star rating system as more of a reminder that I've seen this profile before, and of course I give all of my real life friends five stars.
Dross fucked around with this message at Apr 14, 2011 around 13:38 |
| # ? Apr 14, 2011 13:36 |
|
|
aliceamadee posted:The best is "Is the earth bigger than the sun?" haha I actually prefer the one featured on the "hatethefuture" blog: 
|
| # ? Apr 14, 2011 13:41 |
|
|
Fnoigy posted:I think it's because if they included all that, it wouldn't be a fun article so much as it would be something out of a scientific journal, which nobody likes to read. The guy doesn't claim perfect causation, just correlation, and he presents it all in a way that the common person can understand AND find interesting. Mainly I'm annoyed because all of my non stats friends send me links to it (because I'm a statistician) and go, "isn't it so cool what they are doing with stats?!?" They see a plot with something going up and they assume that it means x causes y. I just wish there could be a bit of discussion about uncertainty on his posts. It is basically posting 50 versions of the number of pirates vs. global world temperature plot and making all these suggestive pokes about how interesting it all is.
|
| # ? Apr 14, 2011 13:49 |
|
|
Chin Strap posted:Mainly I'm annoyed because all of my non stats friends send me links to it (because I'm a statistician) and go, "isn't it so cool what they are doing with stats?!?" They see a plot with something going up and they assume that it means x causes y. I just wish there could be a bit of discussion about uncertainty on his posts. It is basically posting 50 versions of the number of pirates vs. global world temperature plot and making all these suggestive pokes about how interesting it all is.
|
| # ? Apr 14, 2011 14:14 |
|
|
Chin Strap posted:There isn't a single measure of standard error on anything he posts. "This number is bigger than this other number" doesn't help you when you don't know error bars on those numbers. And how much is looked at that is subsequently thrown away because it doesn't look like anything? I could make 1000 random plots of noise, find 5 that look like they have a sort of signal, and make a blog post about it. Doesn't mean that it is interesting or has any sort of statistical rigor. Surely you're aware then that standard error estimations are pretty much a non-issue with such a large sample? It this point, EVERYTHING is statistically significant (even if he isn't doing hypothesis testing): the important consideration is whether or not the difference in means is something to care about. For example, lets say he found that 50.1% of females and 50% of males had some sort of trait, with a standard deviation of 20 each and n of 1.5 million (or .75m per group). The difference between men and women is significant (2-tailed) to p = 0.002 (you can check it yourself). Even with crazy conservative multiple comparison corrections, all of his work is still very significant. It doesn't matter how many analyses he's done when he's finding differences of 5-10%, much less 0.1%. But all of that is a irrelevant anyways, because he is only using descriptive statistics, not inferential statistics. He's showing people data that is likely to be "true" (in the sense that his sample actually reflects the population), interesting, and easily understandable.
|
| # ? Apr 14, 2011 14:37 |
|
|
Forever_Peace posted:
The populations aren't uniform, there are weird sampling biases, I can go on and on and on. The freshman two sample independent normal calculation you threw up isn't at all applicable to this (or basically any) real world situation. There are plenty of ways he could show your hypothesis to be true if it is. He doesn't. I'm fine leaving it at "here's some funny numbers" I just wish he would be more explicit that it isn't a good statistical handling.
|
| # ? Apr 14, 2011 14:46 |
|
|
Isn't getting a phd in stats enough that you don't also have to jerk yourself off in this thread.
|
|
| # ? Apr 14, 2011 15:01 |
|
|
aliceamadee posted:I think that's a big reason why the quiver/ quickmatch/ homepage doesn't take physical characteristics into account at all. Forever_Peace posted:But all of that is a irrelevant anyways, because he is only using descriptive statistics, not inferential statistics. He's showing people data that is likely to be "true" (in the sense that his sample actually reflects the population), interesting, and easily understandable.
|
| # ? Apr 14, 2011 17:05 |
|
|
|
| # ? May 19, 2013 18:50 |
|
|
Hi why did they remove subjects? I was literally on the site when it happened, I refreshed my mailbox and they were gone. At first I was annoyed. I generally send messages with subjects better than "Hey" "sup" "hi" etc. I like to think that it makes a difference. From talking to some female friends of mine however, they all say that they read almost every single message, even if the subject sucks. That makes me think it's useless, and good riddance. Coming up with a good subject can be hard. Thoughts/musings/reasoning? I've come around to it - but I also had a problem where a girl who I was planning on meeting for coffee this weekend lost my message without the subject. I had changed my picture in the interim, and she didn't realize it was me. Luckily she got in contact with me today (forgot we had exchanged numbers), but I think this would have been avoided with a subject.
|
| # ? Apr 14, 2011 18:22 |
|
