Some more thoughtsnaturalbornJuly 14 2003, 17:29:43 UTC
The superficially fast looking technique I suggest actually has serious performance problems under some circumstances. Consider the case where A certs only B, B certs (A, C), C certs (A, D), D certs (A, E), etc. Y is the seed. How long will it take to add Z to the list? I'm not sure if there's a trick to efficiently compute it.
The weakening problem of certing more I mentioned before is quite common and probably serious. Nodes which only get to cert a single other node into membership will likely cert noone into membership if they cert two other people, so it creates a very strong and artificial disincentive for certifying, which is a bad thing.
Applied to people (the LJ friends-list example) one issue immediately asserts itself: you're assuming zero impedance on each arc. A better model IMO would (keeping the water metaphor) place a valve midway along each arc, reflecting the trust-receptivity of that particular arc. The greater-rank nodes in the current algorithm are interestingly those holding greater trust *or distrust*... the closer-in nodes have greater variability, perhaps linked to more exposure or knowledge between nodes. Some arcs are completely open, others are completely blocked -- it picks up enemies as readily as friends. Lesser-rank nodes are probably less-connected or well known, and hence assuming a fixed valve setting is OK.
I wounder if it wouldn't make a better algorithm if more water would flow dependig on how many friends in common do you have with a particular person. AFter all if we have many friends and communities in common chances are that we are very strongly linked, and possibly his other friends might be more of interest for me.
I radically changed my friend list, to make a search using only my closed friends. I probably pissed off also quite some people in the process, so I'll have to go back asap. In the meantime, could you please tell me when I can try again? For now I am always getting the same results.
Comments 19
The weakening problem of certing more I mentioned before is quite common and probably serious. Nodes which only get to cert a single other node into membership will likely cert noone into membership if they cert two other people, so it creates a very strong and artificial disincentive for certifying, which is a bad thing.
Reply
Reply
Reply
Reply
I radically changed my friend list, to make a search using only my closed friends.
I probably pissed off also quite some people in the process, so I'll have to go back asap.
In the meantime, could you please tell me when I can try again? For now I am always getting the same results.
Thank you,
Pietro
Reply
Leave a comment