Insights about tinder and online dating (long thread)
About a year ago some friends scraped 70k pics on tinder and it was reported a few months later: https://bit.ly/2ZbFVwf They said their goal was to outdo the guy that did 40k and was reported in the news a few years ago. Anyways, not to be outdone, I decided to scrape millions and do 2 TB worth of data. I gained several insights. One thing I noticed is that as I was scraping male and female profiles at the same rate, I was always running out of new female profiles to scrape whereas I never ran out of males. I realized that tinder user base is like 99% male even though their official numbers say otherwise. Maybe this is why the FTC is suing match group for making fake female profiles? https://www.ftc.gov/news-events/press-releases/2019/09/ftc-sues-owner-online-dating-service-matchcom-using-fake-love After scraping about 250k female profiles I basically couldn't scrape much more without running into massive duplicates in the major cities (bay area, Seattle, new York, Chicago, LA, San Diego). I'm sure there's lots of churn but there are also tons of catfishes as well. Elo hell:π₯ Sometimes if an accounts ranking falls too low it will fall into what I call ELO hell, where it basically returns the Same recommendations continuously ad nauseum. Once this happens to an account there's really nothing you can do other than delete it. I estimate there are millions of males stuck in elo hell so basically from tinders perspective all they are good for is ad revenue (lol). This is similar to getting shadow banned, which can easily happen as well. Free gold:π Tinder will give you free gold under certain circumstances. I believe it's something like getting 10 superlikes in a row from high ranked players. I used a teen bikini pic in the bay area and got it within about 20 minutes. It's very easy for girls, however for guys the only time when I got free gold was when I used a model and said I had a private jet and went to Harvard. I tried replicating this many times but in general the results were hit or miss. Also keep in mind that if you change your pics after it gives you gold, it will rerank you (probably at the bottom too). They do use aws Rekognition to detect duplicate photos but I don't think they do a very good job because of how easy it is to bypass. Authentication:π Authentication to the tinder API requires a token (duh), so one way to get this token is to have a service running on your phone/ phones (remember they are against VoIP numbers) that will send the numeric text code you get to your main server (i.e 123456) so that you can continuously get new tokens when they expire (every 24 hours or whatever). I bought lots of sim cards from twilio to get around the VoIP restriction but I'm sure there's a cheaper way. GPS:π°οΈ if tinder can't reach your GPS, it will use your last known GPS connection, so you can just set this once to wherever you want when you first create and account (spoof GPS is easy a long with using an IP address from where you want) Bumble authentication: For bumble you can just replay your requests (saved with your cookie) so you don't have to reauthenticate each time to make API calls. Bumble was acquired / merged with Badoo, and when you look at their web requests you can see they just copy and pasted the API endpoints , and interestingly enough they also left a web beacon that says anyone that sees it should apply for a job in their site. π€. About rate limiting: this has changed a lot over the past year but the basic idea here is that they rate limit mainly on ID, so IP addresses are not a big deal. My last check it was about 50k API calls within 24 hours would rate limit but I'm sure things have changed since then. Bikini pics: π One thing I noticed about Americans vs non Americans is that Americans tend to flaunt their "assets" a lot more than others. I spent 2k in AWS credits to filter down the female pics to just the bikini pics (so now I have 164k bikini pics, it was about 10%). My goal was to use StyleGAN2 to be able to generate more content, however the compute time and technology is not there yet (mode collapse issues, resolution issues, not squarez black and white, too complex, etc) Flagging easy lays: ποΈ One thing I'm sure about is that Tinder has some way of flagging "easy lays" or whatever because they need to spread those around to keep the drip feed going (males that spend a lot of money need to get an occasional lay to keep spending lots of money). I'm not sure how their model does this other than doing some analysis on phone number exchange or conversation analysis. I noticed a lot of bots use Unicode encoding for their text (i.e. instead of a they will use Γ for instance), not sure if tinder accounts for that in their spam algorithms. They do outsource some (or maybe even all) of that to Arkose labs. Just like bumble, when in Tinder a female gives a male a phone number after a conversation that male will get a huge boost in the rankings. It is in their best interest to perpetuate hook up culture and hypergamy so that the supply of females is even more restricted and males pay even more money. There are a lot of studies done in this which I won't get into here, but if you are spending money on a dating app you are throwing your money in the trash and it's much better spent in other ways. Note that I do not use online dating and did not do anything to be malicious, deleted all the data months after I scraped it, and such. I was just curious about how certain technology works. βΊοΈ
GizmodoDid you analyze differences based on race?
No, how would I do that? Classify race using aws Rekognition? If you want data on race there's a lot of blackpills in the archived posts by Christian rudder, founder of OkCupid. In fact the company got so concerned they were bad for business they expunged the blog posts and republished the book
Very nice! By the way if you do the same thing for seekingarrangement pretty sure the male/female ratio will be reversed and you get more bikini pic ;)
I'm not sure about that. It seems seeking arrangement operates on the same principle as Ashley Madison - aka make fake females that message or at least "view" males to get them to sign up for premium
Nice!
Why are 70k pics a big deal? I remember in 2015 I ran a simple script in C# that was sending 'pass' requests and downloading pics. After a few hours it downloaded 14k. I don't know if anything changed now, but I think anyone can download 70k pics with 30 lines of code even today.
Because the media doesn't understand technology or have any clue what an API is To scrape 2tb I had to have about 10 accounts going simultaneously over several weeks, it would've gone faster but originally my computer wasnt using an SSD which was causing a lot of lag, also my scraper program had problems and wasn't optimized. Switching from Ruby to golang and to an SSD helped a lot
I would expect that a high number of requests with the same token would result in a ban. But if they don't ban people, then internet+ssd speed is the only limit.
That was some good research.
Interesting, what was your motivation behind this work? Is Lyft not keeping you busy?
What place in the US would be the best place to find dates for a straight male?
Probably NYC
"Just like bumble, when in Tinder a female gives a male a phone number after a conversation that male will get a huge boost in the rankings." You know this how?
Wonder if you can now use the profile pics youβve gathered to do some experiments. Perhaps same profile in different cities to maybe show if people in LA really are more vain than people in other cities.
well theres been experiments where they use a male model and write he is a child molester in the profile and women still want to hang out with him. So I'm pretty sure at the end of the day most people don't read profiles and just look at pics.
Thank you for all this effort! What advice do you have for guys who are struggling to get matches or are getting ghosted in conversations?
Be tall. The average guy is inflating their height by 3 inches online. In person, a lot of guys are wearing boosts in their shoes. Also be good looking, be rich. If you have a trust fund and are working on passion projects that's a lot more attractive than working for a manager in a big Corp. To be good looking, you can get jaw surgery, liposuction, permanent makeup, tricep implants, pallete expansion, modified lefort 3, leg lengthening, skin resurfacing, lots of stuff. If you drive a nice car that's a big plus. But for the love of God don't listen to people that say you just need to make her laugh. There's a million studies I could link but here's one:https://www.sciencedirect.com/science/article/abs/pii/S0191886912005028?via%3Dihub Attractive males are considered more funny, but funny males are not considered more attractive
Lol, you want dating advice? Donβt ask a guy who has the time to do a project like this. Ask the guy who gets laid 3x a week. Short version is this: be fit and be genuine. You need to nail one of 3: money, looks or humor while being good or average in the other 2. Lastly, be insanely confident( but be genuine), selling the vision that you will one day make is the next best thing.