3 ms·

>I don't see any problem in using ELO. >it makes sense to show most profiles in similar normalized ELO ranges Let me highlight (but not explicitly point out,

>I don't see any problem in using ELO. >it makes sense to show most profiles in similar normalized ELO ranges Let me highlight (but not explicitly point out, so you may mentally connect the dots yourself) the exact problem with both of these statements: Suppose the 'logic' of whatever meta-algorithm one implicitly presupposes as behind picking the desiderata for picking the appropriate algorithm to use for the problem as principally sound. Then, by extension WOLG, it NOWADAYS actually makes no sense at all to continue using ELO, because significantly better algorithms have long since come along. Two by now both ancient examples, check the first one for why & how ELO sucks even for its original intended usage, and check the second one for a tremendously improved version: * https://www.microsoft.com/en-us/research/project/trueskill-ranking-system/ https://www.microsoft.com/en-us/research/project/trueskill-r... (2005+) * https://www.microsoft.com/en-us/research/publication/trueskill-2-improved-bayesian-skill-rating-system/ https://www.microsoft.com/en-us/research/publication/trueski... (2018) (Citations: https://scholar.google.com/scholar?cites=10179469592728174863 https://scholar.google.com/scholar?cites=1017946959272817486...) Oh and, as one may guess, neither of these examples represents the State of the Art in this sector (Check the citations above for that.). Also, for more papers by the Microsoft Research group working on that project, check here: https://www.microsoft.com/en-us/research/project/trueskill-ranking-system/#!publications https://www.microsoft.com/en-us/research/project/trueskill-r... Now that we've clarified that, I'd like to ask: Do you still feel sure whether this seems like an appropriate class of algorithms to leverage for dating? Assuming of course that users want to find partners, as opposed to good opponents. Albeit: One might speculate algorithmically supplying well-matching opponents instead of potential partners, while (mis)labeling them as more potentially likely compatible than by chance partners would make for a tremendously better—albeit highly amoral—repeat business model on part of the platforms, at least in the relative short-term, wouldn't you say?

actuator 6y ago
If your argument is that there are better 'rating' algorithms than ELO which can be used. Sure, go ahead and use them. I wasn't championing ELO specifically. I meant any rating algorithm which can theoretically rank individuals. Obviously, no solution of ranking individuals on the concept of wins/losses is going to be perfect. Btw, I had the luck of implementing the bayesian Trueskill system you mentioned in some product I worked on. > Do you still feel sure whether this seems like an appropriate class of algorithms to leverage for dating? Assuming of course that users want to find partners, as opposed to good opponents. Imagine there being `n` heterosexual males and `n` heterosexual females on a dating platform. If all have to be paired as couples and if we can rank them by their 'popularity'. Don't you think most at the first glance would be more comfortable picking someone about their rating or higher? Just look at the skew of likes people get depending on their attractiveness from old OkCupid blogs. In dating apps, the rating most likely comes from the person's looks. Does that mean they are good matches? No way of knowing this. But does that mean they are more likely to swipe on each other? Probably yes. We can show them people at random but that doesn't mean that they will swipe on someone whom they don't consider to be at least in the same ballpark as them on attractiveness. The only way we can get better potential partners is if we have a lot of potential data about every user not limited to just looks. Also, even in that case if the other person is not on the same scale on attractiveness, the other person will have to take a leap of faith on whatever matching system they have.(which if you go by social experiments, people don't do) Considering how much data FB has about users through their platforms, they are the ones who truly can come close to matching people on common interests but I am not sure how successful even their dating venture has been.

actuator 6y ago

If your argument is that there are better 'rating' algorithms than ELO which can be used. Sure, go ahead and use them. I wasn't championing ELO specifically. I meant any rating algorithm which can theoretically rank individuals. Obviously, no solution of ranking individuals on the concept of wins/losses is going to be perfect. Btw, I had the luck of implementing the bayesian Trueskill system you mentioned in some product I worked on. > Do you still feel sure whether this seems like an appropriate class of algorithms to leverage for dating? Assuming of course that users want to find partners, as opposed to good opponents. Imagine there being `n` heterosexual males and `n` heterosexual females on a dating platform. If all have to be paired as couples and if we can rank them by their 'popularity'. Don't you think most at the first glance would be more comfortable picking someone about their rating or higher? Just look at the skew of likes people get depending on their attractiveness from old OkCupid blogs. In dating apps, the rating most likely comes from the person's looks. Does that mean they are good matches? No way of knowing this. But does that mean they are more likely to swipe on each other? Probably yes. We can show them people at random but that doesn't mean that they will swipe on someone whom they don't consider to be at least in the same ballpark as them on attractiveness. The only way we can get better potential partners is if we have a lot of potential data about every user not limited to just looks. Also, even in that case if the other person is not on the same scale on attractiveness, the other person will have to take a leap of faith on whatever matching system they have.(which if you go by social experiments, people don't do) Considering how much data FB has about users through their platforms, they are the ones who truly can come close to matching people on common interests but I am not sure how successful even their dating venture has been.