Information Technology Reference
In-Depth Information
quite
middle
good
0
1
2
Figure 11.4
Fuzzy Intervals. (Reproduced with kind permission of Springer Science+Business Media
2003)
C
changes in big (positive or negative) values have more impact on the FCM, this is a tolerable
result even if it does not correspond with a general cognitive model.
11.10 Description of the Model
Even if FCMs are graphs, ours can be seen as having four layers. The first layer models the
influence of the 'beliefs sources':
Direct Experience
(e.g. 'In my experience
...
'),
Catego-
rization
(e.g. 'Usually doctors
...
'),
Reasoning
(e.g. 'I can infer that
...
'),
Reputation
(e.g.
'A friend says that
'). Their value is meant to be stable (i.e. it does not change during
computation), because these nodes could be assumed as being the result of an 'inner FCM'
where each single belief is represented (e.g. direct experience about ability results from many
nodes like: 'I was visited
many times
by this doctor and he was
really good
at his work', '
Once
he made a
wrong
diagnosis',
...
). So their value not only represents the strength of the feature
expressed in the related beliefs, but also their number and their perceived importance, because
belief sources represent the synthesis of many beliefs.
The second layer shows the five relevant basic beliefs:
Ability
,
Accessibility
,
Harmful-
ness
,
Opportunities
and
Danger
. These basic beliefs are distinguished in the third layer into
Internal Factors
and
External Factors
.
Ability
,
Accessibility
and
Harmfulness
are classified as
Internal Factors
;
Opportunities
and
Danger
are classified as
External Factors
.
Internal
and
External Factors
both influence
Trustfulness
, which is the only node in the fourth layer. For
the sake of simplicity no crossing-layer edges are used, but this could be easily done since
FCM can compute cycles and feedback, too.
...
11.11 Running the Model
Once the initial values for the first layer (i.e. belief sources) are set, the FCM starts running.
The state of a node
N
at each step
s
is computed taking the sum of all the inputs, i.e., the
current values at step
s-1
of nodes with edges coming into
N
multiplied by the corresponding
edge weights
. The value is then
squashed
(into the -1,1 interval) using a
threshold function
.
The FCM run ends when an
equilibrium
is reached, i.e., when the state of all nodes at step
s
is the same as that at step
s-1
.
Search WWH ::
Custom Search