Geoscience Reference
In-Depth Information
User interface
Expert advice
Facts/databases
Inference engine
Knowledge base
FIGURE 11.1
The basic components of an ES.
an automatic way for the user to enter knowledge into the system rather than having the knowledge
engineer explicitly code the knowledge. Knowledge (in the knowledge base) may be represented
in several ways including rules
(also referred to as production rules), semantic nets, frames, script
logic, conceptual graphs (Turban 1995; Jackson 1999; Giarratano and Riley 2005) and ontologies
(Fonseca et al. 2002; Albrecht et al. 2008; Couclelis 2010). The choice of an appropriate method for
storing this knowledge depends on the nature of the problem concerned (Moore 2000). In particular,
it depends on the pre-existing format of the knowledge, the types of decision that the system will
take and the inference process used in order to make these decisions (Medsker and Liebowitz 1994).
Rules are the most popular representation technique used in ES (Medsker and Liebowitz 1994;
Giarratano and Riley 2005). This is because rules are a very simple form of representation that
is very close to the cognitive behaviour of people when they make decisions in many cases, both
simple and complex. Moreover, rules provide a flexible way of building knowledge. In particular,
knowledge can be built incrementally, that is, rules in a knowledge base can be gradually added
once a new piece of knowledge emerges. In this manner, the performance and the validity of the
system can be continually verified (Giarratano and Riley 2005). Moreover, rules facilitate inference
and explanations, modifications and maintenance as well as the incorporation of uncertainty. Since
complex problem domains may require hundreds or thousands of rules, this may create problems in
both using and maintaining the system.
An ES that uses rules is also called a rule-based or production system. Rules were developed by
Newell and Simon for their model of human cognition (Turban 1995). The basic idea is that knowl-
edge is represented in the form of condition-action pairs, for example,
IF
this condition (or premise or antecedent) occurs
THEN
some action (or results or conclusion, or
consequence) will or should occur
.
Inferencing with rules is carried out based on two search mechanisms: forward chaining and back-
ward chaining. Forward chaining is a data- or event-driven approach that is used when data or basic
ideas or events are the starting point. In other words, it is a bottom-up reasoning approach, since it rea-
sons from low-level evidence, that is, facts, to the top-level conclusions which are based on these facts.
Thus, the system does not start with any particular goal(s), but it works through the facts to arrive
at conclusions. This approach is usually used for data analysis and in the design and formulation of
concepts (Medsker and Liebowitz 1994). Moore (2000) notes that this method is appropriate for repre-
senting
what if
scenarios, which are a common modelling approach in geospatial planning processes.
In contrast, backward chaining is a goal-driven approach which starts with a goal (i.e. a hypotheti-
cal solution) and searches for rules that will provide the evidence to support this goal. This approach
