Similar-interests-listings aren't symmetric

[jay]

A minor quirk (or feature?) of LJ's find-users-by-similar-interests is that scoring is not symmetric. A's score on B's similar-list does not necessarily equal B's score on A's list. Given the same shared interests between B and A, this is counterintuitive to me. For example, compare myself with cjsmith, who is probably headed home now from Houston... on her similar-interests list, the LJ algorithm scores me at 18.126, 3rd on her list. But on my list, CJ only gets 13.179 and 90th place. For the same set of shared interests... weird.

Off to bed now...

Comments

[technomom]

That's odd. My "magic index" with sambear is 35.593. The second person on my similarity list, though, is the_curmudgn. My magic index with him is 21.588, but his with me is only 16.738. I wonder how the formula works?

Different scores

[oakdragon.livejournal.com]

Offhand, I expect that the score is not based upon how many similar interests you have, but on what percentage of your interests are matched by the other. She has 30 interestes listed while you have 89. A hit on her interest list is probably worth 3.33%, whereas a single hit on your larger list might count for 1.12% I only have 5 interests listed currently, so a single match would probably give a 20% score.

Re: Different scores

[treacle-well.livejournal.com]

Yes, I think that's sorta how it works, only some weight is also added in for interests that are shared by only a small number of LJers. That bit isn't relevant to the original observation though, since if those "rare" interests appear on two people's interests list, they would have the extra weight on both sides.

[cjsmith.livejournal.com]

I put flying but didn't put aviation. Should fix that? Would mess up the numbers. Plus am too lazy right now... :-)