8 ms·

It's a spectrum, there's no black-and-white definition. On one end there are posts that get votes from people who visit the front/new page, vote for multiple th

by tlb 12y ago

It's a spectrum, there's no black-and-white definition. On one end there are posts that get votes from people who visit the front/new page, vote for multiple things and make comments that get upvoted, and on the other end there are posts that get direct visits from accounts that only ever view and vote for posts from that one author. In the middle, there's a weighting function with empirically tuned parameters. While there are a few clever voting rings, the vast majority are just one guy with many sockpuppet accounts upvoting self-serving crap. HN doesn't need anything as fancy as HyperLogLog. Most voting rings are apparent within the first 20 votes so elementary data structures work fine. (I know how the HN system works and anecdotally how Reddit's used to work, but I'm being a bit generic here to avoid revealing that One Weird Trick to Get to the Top of HN.)

cyphunk 12y ago
if a ring introduces noise (meaning they attempt to avoid the "not clever" trap you describe) how does this effect the measurement? I can assume that introduction of noise is somewhat counter to the greed and objective of the ring leader. But now the objective would change to be meerly introducing enough noise to putting the metric on this ring in a state of question. With that though, of course this is better than a model that requires no noise. Forcing noise from the edges reduces the set of entries requiring human review and, still, human review might reveal with sufficient weight the existence of a possible ring. But, this is an evolution to the cat+mouse stage. Hence, rings could still survive if clever? With that question (ring survival with measured noise) set aside, it seems if the cat+mouse were to come in earnest that reputation dependent models would be a driver for mapping user "types" (social, political, behavioral models) in a much more accurate manner than the current driver of advertising (google wanting to increase price of adds)

by tlb 12y ago

cyphunk 12y ago

if a ring introduces noise (meaning they attempt to avoid the "not clever" trap you describe) how does this effect the measurement? I can assume that introduction of noise is somewhat counter to the greed and objective of the ring leader. But now the objective would change to be meerly introducing enough noise to putting the metric on this ring in a state of question. With that though, of course this is better than a model that requires no noise. Forcing noise from the edges reduces the set of entries requiring human review and, still, human review might reveal with sufficient weight the existence of a possible ring. But, this is an evolution to the cat+mouse stage. Hence, rings could still survive if clever? With that question (ring survival with measured noise) set aside, it seems if the cat+mouse were to come in earnest that reputation dependent models would be a driver for mapping user "types" (social, political, behavioral models) in a much more accurate manner than the current driver of advertising (google wanting to increase price of adds)