[foaf-dev] FOAF-based whitelisting project

elw at stderr.org elw at stderr.org
Mon Mar 12 14:36:58 UTC 2007



> My idea is that we use a single trust metric for everything, I would
> suggest an integer in the interval [-100,100], -100 indicating strong
> distrust, 100 indicating full trust. That metric could be returned from
> either a web service or a locally hooked up system, but I think that it

A unidimensional trust metric is often not particularly useful.  [This was 
something I struggled with mightily in MA thesis work, some five years 
ago, looking at how then-current moderation systems were using trust 
metrics.]

Better would be to define terms that allow you to specify terms and 
degrees of trust, as relates to particular domains of discourse or domains 
of agreement, such that I can trust your opinions or attitudes in domain 
#1 but choose to think you're full of hoo-ah in domains #2, #3, #4.  I 
think this is a better fit with FOAF/RDF/SW technologies in general, 
though it does demand a little more definition of vocabulary terms....


> I think we need to think ahead in terms of sufficient performance. Since 
> there are allready like 17 million FOAF profiles, and we know that every 
> received email will result in a lookup, we know as a benchmark, we need 
> to have decent response times given those constraints. It should not be 
> bandwidth intensive (I'm thinking a REST URL, responding with a single 
> text/plain containing the metric), but computationally, it may require 
> some heavy lifting.


Yes, computational demands for this sort of thing will be high.  In my 
experience the best way to think about trust is in relative rather than 
absolute terms -- so that, for example, you aren't trying to compute paths 
across a 17M node trust graph.  Localized trust - such as within a 
community or within a bounded range of users - is much less 
computationally expensive and can be quite effective.


[There ought to be substantially more than 17M foaf profiles - LJ alone 
has 12M, plus sites like LiveDoor (7M?) and others in the non-english 
space...]


> My current idea is simply to gather all the FOAF we trust, including 
> vCards and whatever, do a little reasoning to produce a weighted graph 
> which can be quickly queried. The main thing with foafwhite.rb is to 
> offload this logic to each user, which is an interesting idea, but what 
> I had in mind was to centralise it.

There's a heck of a lot of FOAF out there.  ;-)  Centralization efforts 
fall apart fairly quickly, in my experience - we've crunched foaf on a 
quad-cpu ultrasparc box with 32GB of memory.... *grind, grind, grind* go 
the hard discs into swap death.....

--e



More information about the foaf-dev mailing list