Analyzing Tinder's Algorithm and RSRR

I reverse engineered Tinder to figure out more about their algorithm and how they

get eight point five million people to pay every month.

I discovered that Tinder's calculating two things, what you want and how much

you're wanted. Tinder learns your type by turning every

photo you swipe right on

into an embedding and averaging them.

It even times how long you stare at each photo.

Tinder also tracks how often people swipe right on you.

They call it RSRR, your Right Swipe Received Ratio, which

is basically out of

everyone who sees you, how many actually want you?

Tinder even trained a model that runs on your phone

and can predict that number for

every one of your photos. Their photo selector feature scans up to

twenty thousand

photos in the last two years, and they run this model on every photo of

you and

pick the best twenty-seven. Tinder claims everything stays on device,

but they

still collect each photo's RSRR score, plus how explicit it is and how young you

look in it. So I pulled the model out and ran it against my own synthetic data

set

to get a window into their algorithm.

I discovered that across the board, mirror selfies triple your score, and more

revealing clothes pushes it even higher for both men and women.

However, the darker your skin, the lower your score.

For men specifically, sunglasses and outdoor photos help

while doing nothing or

worse for women. For women, being heavier and older does hurt your

score

significantly, while neither does anything to the men's

score.

Then Tinder uses this RSRR score to sell you.

If Tinder decides your profile is popular, some of the people who swipe right on

you will enter a paywall flow with the sale literally tagged as high RSRR

upsell, so Tinder can measure exactly what your

popularity earns.

Let me know in the comments what content you'd like to see next.