[MUD-Dev] Trust systems and Player-Run Reputation

[Travis Nixon]

From: <Daniel.Harman at barclayscapital.com>
>> -----Original Message-----
>> From: Brian Hook [mailto:bwh at wksoftware.com]

>> The biggest problem is normalizing the scale since this is not a
>> clamped rating system.

> Maybe I've missed something, but the biggest problem I see with
> this non-directed cyclic graph, is that it is exactly that.

> You end up with an potentially unstable system, which even if it
> is stable, would require many iterations to arrive at a stable
> point.

> (If its not clear what I'm driving at, consider that my rating of
> Boffo is based on my friends rating of him, but my friends rating
> is based on mine.  Thats a two way dependency which is a problem,
> then start adding friends in and it quickly becomes exponentially
> harder).

> Of course I may have missed something here?

I dunno.  Forget the whole friends thing though.  Friends are
irrelevant, at least in the original version.  Your rating of Boffo
(assuming you haven't given him one yourself) is not based on your
friends' rating of him.  They're based on the ratings of other
people who rate people the same way you do, and have rated Boffo.
Boy, I'm not making this any clearer, am I.  :)

Quick example.  We have players A, B, C, and D.  D is a jerk.  C is
a friend of D.  Everybody has met everybody except for A and C.  A
knows that D is a jerk, B knows that D and C are both jerks.  So,
our ratings are as follows:

  A has rated D as evil and B as good
  B has rated D as evil, C as evil, and A as good
  C has rated D as good and B as evil
  D has rated C as good, B as evil, and A as evil

Ok, A encounters C.  They have never met before, A has not
personally rated C.  If he has, that's the rating he sees.  So the
derived rating for C is evil, based on B and D's ratings.  B's
rating counts for everything, because A and B rated D the same, so
their opinions match in this regard.  D's rating counts for nothing,
because D rated B as evil, while A rated B as good.  And C's rating
obviously doesn't count, because you don't rate yourself.

In this example, A and B are probably friends, despite what I said
earlier about forgetting friends.  That's not because they have to
be friends for the system to work, though.  That's just because I've
about stretched the limit of connections between 4 people, and
because A and B have to have an opinion of each other for it to work
with so few.  :) With more people, you'd have matching opinions
without the people whose opinions match necessarily having ever met
each other, which is the whole point.

It's not a cyclic graph, because not every rating is derived, and
because ratings that are derived are based on ratings that people
actually set, which is why the whole system would be useless if
people didn't use it.  If nobody ever rates anybody else, everything
falls apart.  You have 2 types of ratings.  Set and Derived.  When I
meet Boffo and give him an evil rating, that's a set rating.  If
I've never met Boffo, then the only rating I have of him is a
Derived one, based on other people's Set ratings.  Specifically,
based on the Set ratings of other people who have similar Set
ratings to mine about other people like Bubba, who I have met and do
have a Set rating for.

Am I making anything even remotely resembling sense yet?  I can see
the picture and all the relations in my head, but I don't seem to be
putting it to words very well. :)

_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev