Re: META: personal server-side filtering

Christopher Maloney (
Sun, 27 Jun 1999 20:13:07 -0400

Eugene Leitl wrote:
> Here's an interesting new filtering variant we've cooked up with
> Cris Rasch today.
> To recapitulate: my original proposal was to establish a pair of
> lists: extropians moderated/unmoderated. Everything rejected from
> extropians moderated would go to extropians unmoderated that anybody
> who subscribed to both lists would get the full feed, if so inclined.
> This scheme is simple to implement, but has obvious flaws. To avoid
> pressing anybody in a single quality metric it should be possible to
> associate each subscriber with a ranking matrix against the other
> participants (which could be inquirable remotely). With 1 k subscribers
> this means keeping track of 1 M short integers, which is tolerable
> nowadays (besides, this can be implemented as sparse matrix). Each
> mail will be attached a pair of urls/listserv commands to
> increment/decrement your individual ranking of the poster this mail
> came from. The matrix gets initialized with zero. All positive values
> including zero mean no filtering. -1 means every second post gets
> through, -2 every third, etc.

I'm not clear on this. What algorithm do you use to determine a person's overall ranking as a function of the 1000 individualspecific rankings? If it's a simple sum, I would think that there might be some interesting failure modes, such as the list becoming dominated by a group who conspire to always vote positively for each other's posts. I don't know what the likelihood of such a failure mode appearing at random would be (in the absence of an overt conspiracy). I think there's a whole science to this sort of thing, but it's not my game.

> This doesn't include topic filtering, but this can be addressed
> separately. To address this, we can agree on an initial keyword pool
> (META:SPACE:GUNS: etc.) and make the listserv let through only
> properly classified messages. If a message is unclassified it gets
> send back with a list of current keywords. There should be a mechanism
> of creating new keywords, and discarding obsolete ones (e.g. by aging).
> I'm fairly certain it would be possible to hack mailman into
> supporting this. Does anybody see any obvious flaws in the scheme?
> ciao,
> 'gene

Chris Maloney

"Knowledge is good"
-- Emil Faber