[expertfinder-dev] Re: Expertise and Trust Vocabs

[Tom Heath]

Hi Axel,

I'm loosing track of different people's posts and replies here (Gmail
isn't as clever interface-wise as it would like to think), so I'm just
going to <snip/> out the most relevant bits... You say:

> anyway, my claim was that the numeric values seem to be arbitrary and
> hard to assess and I was suggesting to start simple.

Exactly the same argument applies to a binary trust relationship -
it's equally arbitrary. At what point does one decide that non-trust
(either zero or null, depending on the situation) is now trust?
Presumably when some certain evidential threshold is reached. However,
any figure chosen as a threshold also becomes fairly arbitrary, so the
issue is the same.

My approach is to infer trust values based on a range of background
data sources, which must be combined in some way. The resulting value
will inevitably be a somewhat arbitrary representation of someone's
trust in someone else (for varying degrees of arbitariness), but being
able to specify some variation in this is immensely important. If
others wanted to take a different approach and simply ask users to
rate 0 | 1 for trust in Person X regarding Subject Y this would be
fine too, and could still be modelled in the Hoonoh ontology as values
of 0 | 1, if you were prepared to ignore the slight hack of modelling
nominal data in what is designed for interval scale.

I think keeping it simple is right, but am inclined to believe that
binary trust is a step too far down the path of simplicity :) Of
course the data I generate will be public (at least that's the plan),
so anyone else will be free to repurpose it into binary trust
relationships, according to their chosen threshold.

Cheers,

Tom.