hey remi,
no "threshold=4" doesn't work for me: i note that on return to a page where i have noted a spammer's diary with a rating of "1", the value is not re-presented to me. perhaps i do not understand how the diary-rating system works (or it's on a 15 min calculation cycle, like the trust metric?)
update: yep, it works. thanks remi. curious: where the heck can i set that? and why isn't the default on recentlog set to very low like... oh... thresh=2?
hi steven, good to hear from you.
the assumption that we made was that all contributions would be of zero or more value, and that there would be nobody stupid enough to endeavour to provide contributions of negative value.
the trust metric keeps people off the front page.
it doesn't keep them off the diary entries.
i'm thinking out loud again.
"negative" certs were discussed many times: i loudly resisted the calls for addition of "negative certs", on the basis that if you haven't got anything good to say, don't say anything.
however, when people _deliberately_ go out of their way to say value-less things, then that's a completely different ballgame.
and advogato is not equipped to deal with that (wrt diary).
here's one possibility to consider:
if an observer's diary _contains_ < img > or < a > tags, then it is simply.... made invisible, and the user is warned that, as an observer, they are not allowed to use < a > or < img >.
also it's important to check for < xxx style="..." > as that can be used to embed images via inline stylesheets.
the reason for making it invisible is that the content pretty much is rendered useless without some convenient markup and images.
i'd far rather we pulled their teeth and made it pointless for peole to post irrelevant content than to encourage people to make "negative" assertions about the content itself.
the "bad neighbourhood" idea - sounds scarily complex, and also sounds like a rewrite of mod_virgule into python is sorely needed (which will reduce the amount of code by approximately 70% by the way).
it sounds to me like the "bad neighbourhood" idea requires "consensus". i.e. if there are less than a certain number of people agreeing that someone is Certified, then that person is "tainted".
i really think it's time to rewrite mod_virgule - especially as the ford-fulkersson depth-first algorithm means that you have to load eeevvveerything into memory to do the trust metric calculation.
if you use a breadth-first algorithm, then 1) you don't have to load the entire graph into memory 2) you can automatically stop at a certain number of degrees 3) you can also work out "consensus" along the way - namely you can check, at each "depth", that the number of people Certifying each of the users at the new "depth", is greater than the minimum number of required Certs (e.g. 3 for Master; 2 for Apprentice and Journeyer).
the min number required for Master definitely needs to be less than the total number of top-level-seeds, btw!!! unless of course you have different rules for top-level-seeds, saying "a top level seed doesn't need "consensus" - their opinions are implicitly trusted and absolute).
a lot of people would be very pissed at the change. well... tough!
p.s. steven: are there backups of the XML files cos my profile has been truncated to zero twice now: the first time lost about a hundred Certs; the second time lost the entire content of my personal details.