not a beautiful or unique snowflake (nothings) wrote,
not a beautiful or unique snowflake

The other thing I hate about matching things like okcupid comes from the fact that you tend to go "ooh, this person is my 3rd best match", not "ooh, this person is a 95% match for me". The reality is you're going to normalize your list and look at your top N contenders, but they don't build for that.

Then the problem is that my best match may be at 89%, but meanwhile I'm not even in her top, say, 40 matches because she has a ton of matches in the 90%s. So, like, this woman I mentioned in comments about the last post, this woman I was friends with in college and had a crush on, is actually one of my top matches nationwide. But how will she react? Am I one of her top matches? Who knows, since I don't know how I rank to her, just my match percentage.

So the thing I hate that I alluded to earlier is that they only provide these percentage numbers--they don't tell you how you rank from the other person's POV. 'Cause maybe in some sense you're only a good match if you're both highly-ranked (not highly-scored) for each other. (pobig may recognize an aspect of his image matching software here.)

But there's no good reason they couldn't tell you your ranking for the other person (other than that it's more database grinding, especially on okcupid where those scores are particularly fluid).

okcupid really exacerbates this problem, because they try to make their percentage matches "valid" by rescaling them based on how many overlapping questions they're using to match you. If you and someone else overlap with only 300 questions, okcupid's confidence level is only, say, 90%, and thus the highest ranking it'll give you is only 90%, and intermediate match values are (to a first approximation) scaled appropriately. So your match percentages against people who have answered fewer are undervalued; and indeed you can drive your score up in other people's rankings simply by answering more questions so that the confidence value rises.

Now, okcupid's design goal is to make these percentage numbers "meaningful and valid", but this is totally frivolous, since there's no other absolute scale to have them meaningful against. All they really matter is for ranking people against each other; and if in fact they are meaningful and valid in their confidence--they've successfully picked a lower bound on what eventual match score they might compute for you with full information--it still isn't useful for comparing people against each other, since you can't really rank two lower bounds. You either accept the lower bound as a best guess (but it's not a best guess, it's a lower-bound guess), or you spend a bunch of time investigating how much you overlap so you know what the confidence level was and reverse-engineer the actual unreduced match value, and rank based on that. Or really, since that's too much work, you just gripe about it on LJ.

Oh, and to people who asked, I remember now that the reason I created an okcupid account was to browse somebody's profile who had made it only visible to people with okcupid accounts. (I don't remember who, but I just rediscovered that they had that option and remembered that was the cause.)
  • Post a new comment


    default userpic

    Your reply will be screened

    Your IP address will be recorded 

    When you submit the form an invisible reCAPTCHA check will be performed.
    You must follow the Privacy Policy and Google Terms of use.