Re: Single model or framework for reputation/trust

Einar Stefferud wrote:
> This thread tends to turn this into a discussion of trust, in which
> case I strongly suggest that we start by defining trust, before we
> start manipulating values of undefined trust in some kinds of
> mathematical or statistical operations.

For definitions of trust we may ask Ed Gerck, he has a large 
collection of. Generally, I don't pretend to define some trust model: 
it is "up to customer" to choose one. What may be helpful to build 
that model in the open environment of the internet is reputation.
I am operating with reputation trivially defined as an average level 
of criterion (promise) fulfillment. Reputation has to be understood 
as an opinion, because even breaking activity into atomic events we 
still need some entity to judge whether they are fulfilling or not.
(E.g. the promise is: "the letter is not spam", while definition of 
spam is left to the recipient. Your inbox - your rules.)

> Are you concerned about the Qualification of information used in
> your trust evaluations?  And who's determination of quality shall
> be used for each piece of aggregated information?  And then for the
> aggregation?
> In short, how do you define trust, other than as a personal
> feeling. What do you get when you aggregate personal feelings?

The only qualification of an opinion is its source. Aggregating 
opinions from different sources, I mix in my own judgement on 
relative worth of original pieces, so the result is my own opinion. 
That is why I do not distinct between individual entities and 
aggregates in the paper.

> Do you use statistical methods?

No. Statistics and probability assumes that a result becomes clear 
after trial (i.e. coin flipping). In the case of promise fulfillment 
uncertainty is not in the process of achieving result, but in the 
result itself. (E.g. I say 'head' while you say 'most likely, tail'.)
So I use fuzzy logic.

--

	Viktor S. Grishchenko