A Fuzzy Implementation for the Socio-Cognitive Approach to Trust - Trust Theory: A Socio-Cognitive and Computational Model - page 318

also true. In order to achieve this result, we have two different strategies: implicit and explicit
revision .
11.13.1 Implicit Revision
In an implicit revision procedure, the error (i.e. the difference between N and n j ) can be back-
propagated and modify the value of i j . The rationale is that the FCM adjusts the evaluation of
its sources, the impact of their opinions (as mediated by our beliefs); for example, in the case
of a source that systematically lies (gives as output the opposite of the final value of the target
node), step to step its impact will be nearer to -1 .
The revision process has two steps . The first step is the change of the impact factor: a
standard back-propagation algorithm can be used in order to achieve this result. The second
step leads to a feedback of this revised value over the nested FCM: some low-weight edges
are assumed that back-propagate the value until the nested FCM stabilizes. For example, if the
impact factor value was lowered, all the nodes in the nested FCM will be lowered a bit, until
it stabilizes.
This procedure has to be better explained. Since the value of the edge (the impact factor)
represents the validity of the source, this value changes as I have a feedback between an opinion
and my final belief. For example, in evaluating the ability of a doctor, a significant difference
between a value furnished by a source and the final value I assume, means that the source
was not totally valid. This can result from different reasons: the source is not trustworthy,
a misunderstanding, poor information and so on. With regard to these problems the implicit
revision strategy is blind: it revises the value of the impact successively to all the nodes in
the nested FCM, without caring what nodes are responsible for the error and so need to be
changed.
This process adjusts the evaluation of a single belief, but it can be in part shared with other
belief impact evaluations with respect to the same source: some nodes of the nested FCM
apply only to the current situation (e.g. certainty about the source) but others are related to
all the interactions with the same source (e.g. trustfulness about the source). So, this form of
learning from a single episode generalizes for the following episodes. This kind of learning is
non specific. We cansider a better one: explicit revision.
11.13.2 Explicit Revision
The explicit revision consists of the revision of some beliefs about the source; since these
beliefs are part of the inner FCM , the (indirect) result is a modification of the impact. So,
explicit revision means revising the values of some nodes of the inner FCM (or building new
ones); the revised inner FCM computes a different impact value f .
In order to obtain explicit revision, the first important issue is to decide where to operate
in the inner FCM. In some cases it would be useful to insert a new node representing a bad
or good 'past experience' under the 'trustfulness' feature; in this case the value is easily set
according to the usual set of fuzzy labels; an example of such a node can be This time John
was untrustworthy .
Even if it is unrealistic to think that a single revision strategy is universal, there can be
many heuristics: for example, a wrong opinion can be evaluated in different ways if I am sure