How can I use SIOC ontology to exploit information from social
networks and generate collective intelligence?

    First you've got to pick a specific problem to solve.  The big
advantage of SIOC is that it captures more structural metadata from
social networks than a webcrawler does.

    I'm not a big fan of 'link graph" data for 'understanding' sorts of
goals;  between noisiness (particularly the links that ought to be there
that aren't) and the power law issues,  link graphs play a secondary
role in work that I do.

    Personally I like to start w/ generic databases (dbpedia,  freebase)
or megaontologies (MeSH,  ITIS) and use those to build "dilettante
systems" that are focused on particular domain areas.  Then I do IX and
other stuff on unstructured information to build up and refine the
knowledge base.

    An interesting question in "collective intelligence" is are you
looking for wolves or sheep?  I gave up on Slashdot back in '99 because
I was involved in a particularly vicious competition for domain names
that people "on the inside" knew was coming.  There was an article on
the topic that was entirely wrong,  and then there were about 200 posts
by people,  none of whom had any idea of what was really happening.

    This was a situation where "those who talk don't know" and "those
who know don't talk;"  I certainly didn't want to tip off potential
competitors to what I knew about the situation or what I was planning to
do about it.  I suppose I could have gotten an ego boost by mouthing off
about it,  but my post would be buried in hundreds of other posts by
people who (i) have better 'karma' scores,  and (ii) sounds like they
know what they're talking about it.

     Looking at what's average is good if you're trying to "keep trouble
(spam,  griefers) out"...  on the other hand,  often the 'average' is
banal,  boring and wrong.  If you're "looking for trouble" you'll find
it in the outliers,  but you just might find something great there too.

    I've been doing a lot of reading on statistical sampling,  but my
interest is in turning the method inside out.  A salesperson,  for
instance,  isn't doing an experiment to find an accurate conversion
ratio for a list of leads.  He's really trying to sell as much as he can
for a given amount of effort,  so he's got every reason to take a sample
(list of people to call) which is as biased as possible.