Curious what the algorithm would be to measure "desireableness". Computer vision on the photos (i.e. measure facial symmetry?) plus the number of other people who have expressed interest in you?

They aren't doing anything that complicated. A popular profile is attractive and chooses a very small percentage of others. The profile that person chooses gets a much greater weighting of desirability by being chosen by the popular profile.

The popular profile that chooses a high percentage of others is not a real human being and/or is selling sex. A pretty irrelevant signal for desirability. It's quite simple. Just like in the real world in that regard, people want higher signals of why they were chosen and it is accurate to be skeptical of an undiscerning attractive person because it usually is a low signal since they don't actually want you, they want to sell something.

Game the system by being more discerning. This is counterintuitive for people wanting matches by statistical probability as it seems like matching faster and arbitrarily will help, but that behavior ensures being downranked to the doldrums with other actually unattractive people.

I do not think there is any need for trying to figure out how to calculate objective attractiveness. Just use someone’s likes and the value of those likes based on the likes of the person doing the like-ing to infer their relative popularities. Can use frequency and rapidity of messaging as an additional data point also.