Information Technology Reference
In-Depth Information
varied then used for generalized explanation structure and thus gets the
generalized explanation structure of suicide(x) while establishing the explanation
structure of suicide (john). Figure 9.9 demonstrates this process.
suicide(john)
kill(john, john)
weapon(gun1)
hate(john, john)
possess(john, jun1)
depressed (john)
buy(john, gun1)
gun(gun1)
Figure 9.8. Explanation structure of suicide (john)
goal concept
suicide(x)
suicide(a)
R1
x/a
kill(a, a)
kill(x, x)
R2
kill(A, B)
{x/A, x/B}
possess(A, C)
possess(x, C)
weapon( C)
weapon( C)
hate(A, B)
hate(x, x)
hate(A, A)
possess(A, Z)
weapon( Z)
R3
R4
R5
x/A
x/A, Z/C
Z/C
depressed (A)
buy( A, Z)
gun( Z)
depressed (x)
buy( x, Z)
Figure 9.9. Generalization of rule suicide(x)
The generalization of Figure 9.9 is based on regression; the constants should
be variable in the predicates.
Another way to implement EGB is on one hand creating the concrete proof
structure of suicide (john) and on the other hand getting the generalized
explanation structure from suicide (
). The connection is that rules concrete
example selected are just those generalized explanation need. That is to say,
problem space should be searched when establishing concrete example
explanation in order to find the usable rules, while creating generalized
explanation can do with searching using the rules that the training examples used.
x
