On 12/4/06, Sean Gilligan <[EMAIL PROTECTED]> wrote:
One quick thought: It might be good to return a fixed-point value from the validate function. Some validators could use a Bayesian filter (or other technique) that is going to return a probability that a given comment is spam. If may even possible to add (or use a weighted average of) the returned values. If both the Bayesian and the link count filters return reasonably high likelihood of spam, that could have a cumulative effect. Of course, a filter that is "certain" that something is spam could always return "1.000"... You could also have something like: 0 - 0.25 = Publish 0.25 - .75 = Moderate .75+ = Reject (but possibly archive?)
That's an interesting idea and it doesn't really cost us anything to use a float rather than a boolean as the return value. - Dave
Users could even adjust the threshold of their filter based upon how much time they are willing to spend moderating. [A funny aside: A blogger (I think it was Jeff Jarvis) was accused of "censoring" comments about "socialism". He finally realized that the word "socialism" contained the word "Cialis" which is a Viagra-like drug, and was triggering his filter.] Let's see if the mailing list lets /this/ message through... -- Sean Dave wrote: > Currently, we've got a couple of different ways to control comment > spam in Roller. > > * Three levels of blacklist: comments that match blacklist are > marked as spam > o Built in blacklist: based on old unsupported MT blacklist > o Site wide blacklist: global admin manages this blacklist > o Website blacklist: each weblog can define a blacklist > > * Comment moderation: when enabled, comments must be approved by > blog owner > > * CommentAuthentcator: determines if user is allowed to comment > o You can plugin your own by implementing the comment > authenticator interface > o Default authenticator does nothing > o Math Authenticator presents math question, verifies answer > o CAPTCHA authenticator is possible too, but we don't ship one > > * Comment throttle: IP addresses that send rapid-fire comments are > banned > > There are problems with each of those methods and even when combined > they're not enough to control spam. We've discussed other ideas for > comment spam control like forcing long comments into moderation, > rejecting comments with too many links and rejecting comments judged > by Akismet to be spam. Those are all good ideas, but if we start > adding special rules ad hoc, we'll end up with a mess. > > What we need is way for Roller site administrators to define a chain > of comment validators so that we and others can add comment spam > processing rules, which are then treated in a uniform way in the > Roller comment servlet. > > Read the rest here: > http://rollerweblogger.org/wiki/Wiki.jsp?page=Proposal_CommentValidators > > Pease respond with comments here on the list. > > - Dave > >
