Information Technology Reference
In-Depth Information
•
Look for the sought cause concept and store it as the head of all causal paths.
•
Locate all the effect nodes linked to this cause concept and create one different
path per node (the head node of all of these paths will be the sought cause node).
•
If the sought effect node is among the effect nodes, stop the recursive procedure
and proceed to evaluate the different paths connecting the node cause with the
node effect. On the contrary, repeat this same step as many times as suggested.
(Because the number of paths can be high if this factor is not checked, we have
chosen a depth of eight nodes as the maximum connecting the sought node cause
with the sought node effect. This number of nodes or levels can be modified).
•
Once all the possible paths have been located, establish some criteria to order
them on the basis of their relevance.
•
Translate the information included in the three more relevant paths into a suit-
able answer (using the same procedure as in
What
-questions).
In the example presented in figure 17.6, the question
How smoking causes death?
has been made. That question displayed the following causal graph as a result.
The graph example shows several causal paths connecting the nodes smoking
and death. Thus, it seems convenient to establish some criterion to classify them.
Kosko's max-min approach to fuzzy cognitive maps serves to this purpose [5]. In-
spired by his work, we calculate the indirect effect and total effect from the node
smoking
to the node
death
establishing a partially ordered set including all the quan-
tifiers labeling the graph.
The lowest of these values is eventually and the high-
est,
provokes
.
Between them, all the rest are in the order that the set
P
shows:
P
. In this ex-
ample there are five possible paths linking
smoking
and
death
, so the indirect effect
of each path is (the number indicates the node):
=
{
eventually
<
can
<
about85%
<
90%
<
provokes
≤
causes
}
•
I
{
n
1
,
n
2
,
n
3
,
n
4
,
n
6
,
n
7
,
n
9
}
=
min
{
causes, causes, eventually, can, can, provokes
}
=
eventually.
•
I
{
n
1
,
n
2
,
n
3
,
n
4
,
n
5
,
n
8
,
n
9
}
=
min
{
causes, causes, eventually, can, can, can
}
=
eventually.
•
I
{
n
1
,
n
4
,
n
6
,
n
7
,
n
9
}
=
min
{
about 90%, can, can, provokes
}
=
can.
•
I
{
n
1
,
n
4
,
n
5
,
n
8
,
n
9
}
=
min
{
about 90%, can, can, can
}
=
can.
•
I
{
n
1
,
n
9
}
=
min
{
85%
}
=
85%.
