How to avoid building erroneous social network graphs?

Hi Folks,

Recall that a "representative hCard" identifies
the person or
organization that represents the current web page.

Suppose that Alice has a web page on a social network.  The
web page
contains a representative hCard for Alice. On this web page
Alice
invites her friends to add comments to her page.  Bob is one
of her
friends, and he adds this comment:

<p>Hi Alice.  Nice page.  I would like to introduce
you to my friend <a
href="Sally.html"
rel="friend">Sally</a> some
time.</p>

Notice the use of XFN: >> rel="friend"
<<

When a robot application encounters Alice's web page it will
see the
representative hCard for Alice and it will see the
XFN-bearing link.
Here's the relationship that the robot constructs:

   Alice is friends with Sally

But that's completely wrong.  Bob stated the relationship. 
The correct
relationship is:

   Bob is friends with Sally

How would a robot avoid this error?

/Roger


_______________________________________________
microformats-discuss mailing list
microformats-discussmicroformats.org
http://microformats.org/mailman/listinfo/microforma
ts-discuss

On 26 Mar 2008, at 19:32, Costello, Roger L. wrote:
> <p>Hi Alice.  Nice page.  I would like to
introduce you to my  
> friend <a
> href="Sally.html"
rel="friend">Sally</a> some
time.</p>
>
> Notice the use of XFN: >> rel="friend"
<<
>
> When a robot application encounters Alice's web page it
will see the
> representative hCard for Alice and it will see the
XFN-bearing link.
> Here's the relationship that the robot constructs:
>
>    Alice is friends with Sally
>
> But that's completely wrong.  Bob stated the
relationship.  The  
> correct
> relationship is:
>
>    Bob is friends with Sally
>
> How would a robot avoid this error?

The situation you describe there is a publishing problem. If
you're  
graphing social relationships, then the page you're
describing has  
been linked to with rel="me", even though the
content of the page is  
authored by multiple people. A page where other authors can
add rel  
links is unsuitable for use as rel="me" node —
except where the  
content management system disallows the rel/rev attribute,
or XFN  
values thereof.

There's no way for a robot to avoid that parsing error.

Ben
_______________________________________________
microformats-discuss mailing list
microformats-discussmicroformats.org
http://microformats.org/mailman/listinfo/microforma
ts-discuss

On Wed, 2008-03-26 at 15:32 -0400, Costello, Roger L.
wrote:
> Suppose that Alice has a web page on a social network. 
The web page
> contains a representative hCard for Alice. On this web
page Alice
> invites her friends to add comments to her page.  Bob
is one of her
> friends, and he adds this comment:
> 
> <p>Hi Alice.  Nice page.  I would like to
introduce you to my friend
> <a
> href="Sally.html"
rel="friend">Sally</a> some
time.</p>

Should Bob be marking up his comment in this way? its
incorrect usage of
rel,
Its Bob who is saying that the current document has the
relationship of a friend of Sally which is wrong.

I guess it could be avoided if every comment had its own
relative url
eg:

http://someblog.com
/post#comment1
http://someblog.com
/post#comment2
etc...

rel
values express the relationships between two urls...

then if Bobs comment was:

http://someblog.com
/post#comment1

<div id="comment1">
<p>Bob said...</p>
<p>Hi Alice.  Nice page.  I would like to introduce
you to my friend <a
href="Sally.html"
rel="friend">Sally</a> some
time.</p>
</div>

then the relationship would be http://someblog.com
/post#comment1 (Bobs
comment) is a friend of Sally.html (Sally)

Thanks

Martin McEvoy


_______________________________________________
microformats-discuss mailing list
microformats-discussmicroformats.org
http://microformats.org/mailman/listinfo/microforma
ts-discuss

What if you put comments into a <blockquote> or a
<q>.  You could
consider XFN in a <blockquote> or <q> to be of
the person being
quoted.

So something like...

<blockquote>Hi Alice.  Nice page.  I would like to
introduce you to my friend <a
href="Sally.html"
rel="friend">Sally</a> some
time.</blockquote>



-- 
Charles Iliya Krempeaux, B.Sc.
http://ChangeLog.ca/

Vlog Razor... Vlogging News... http://vlograzor.com/
_______________________________________________
microformats-discuss mailing list
microformats-discussmicroformats.org
http://microformats.org/mailman/listinfo/microforma
ts-discuss

On 27 Mar 2008, at 01:02, Charles Iliya Krempeaux wrote:
> What if you put comments into a <blockquote> or a
<q>.  You could
> consider XFN in a <blockquote> or <q> to be
of the person being
> quoted.

That might be a valid parsing rule, I'm unsure. Regardless,
it's  
inappropriate for comments. A comment is a piece of original
content  
on the page, not a quote from another source. It would be
appropriate  
for Trackback or Pingback excerpts, though.

B
_______________________________________________
microformats-discuss mailing list
microformats-discussmicroformats.org
http://microformats.org/mailman/listinfo/microforma
ts-discuss

On Wed, Mar 26, 2008 at 7:32 PM, Costello, Roger L.
<costellomitre.org> wrote:
>  <p>Hi Alice.  Nice page.  I would like to
introduce you to my friend <a
>  href="Sally.html"
rel="friend">Sally</a> some
time.</p>

If a page has rel="me" links then it shouldn't
really be allowing
comments that can attempt to build social graphs.

The simplest way to stop it is to add rel="nofollow" to
any comment
links - this has the effect of negating any XFN values in
the links,
as well as preventing linkspamming and all sort of other
good stuff.

As others have said, this is a publishing issue rather than
a parsing
issue.  A page that is linked to with rel="me", and
then allows
outbound XFN values authored by people who are not the
representative,
is broken.

-Ciaran McNulty
_______________________________________________
microformats-discuss mailing list
microformats-discussmicroformats.org
http://microformats.org/mailman/listinfo/microforma
ts-discuss

On Thu, Mar 27, 2008 at 9:59 AM, Dan Brickley <danbridanbri.org> wrote:
>
>  On 27 Mar 2008, at 07:34, Ciaran McNulty wrote:
>  > The simplest way to stop it is to add rel="nofollow" to any comment
>  > links - this has the effect of negating any XFN
values in the links,
>  > as well as preventing linkspamming and all sort
of other good stuff.
>
>  Where is this interaction specified? Should a
compliant XFN parser not
>  emit any data from elements where it finds
rel=nofollow? Does that
>  extend to all Microformats.org ('big M')
microformats?

It's specified here:
http://microformats.org/wiki/xfn-clarific
ations#me_nofollow_interaction

Actually it looks like it only applies to rel="me" from the looks of
things - I'm not aware of any other uFs that interact with
nofollow.

>  > As others have said, this is a publishing issue
rather than a parsing
>  > issue.  A page that is linked to with rel="me", and then allows
>  > outbound XFN values authored by people who are
not the representative,
>  > is broken.
>
>  Presumably they could *author* the links, but they
just have to bear
>  in mind that (if the claims in the page are ever to be
true) those
>  links describe the person who is the 'primary topic'
(or 'owner') of
>  the page.

Agreed - in that case it becomes an issue of trust as to who
you allow
to author that sort of content.

Any commenting/posting system for 'untrusted' users should
be
considering how to filter user input anyhow - if you're
allowing
completely rich HTML editing for third parties you have to
accept that
they'll do 'bad stuff' sometimes - posting huge images, CSS
to blank
the page, embedded malicious OBJECTs and so on.

I've found the only way to guard against markup abuse that
you've not
thought of is to have a whitelist-based approach to what
content you
allow users to author, i.e. a set list of tags/attributes
that are
acceptable.  Blacklisting known abuse routes has always
ended up as an
exercise in firefighting, in my experience.

-Ciaran McNulty

-Ciaran McNulty
_______________________________________________
microformats-discuss mailing list
microformats-discussmicroformats.org
http://microformats.org/mailman/listinfo/microforma
ts-discuss