Geoscience Reference
In-Depth Information
is appropriate when the solution to the problem concerned involves reaching a unique fact/goal
through inferencing. Forward chaining facilitates breadth-first search, which is a graph-searching
algorithm that begins the search at the root node and explores all the neighbouring nodes. Then, for
each of the nearest nodes, it explores their unexplored neighbouring nodes, and so on until it finds
what it is looking for. In contrast, backward chaining is appropriate for a depth-first search, which
is an algorithm for traversing or searching a decision tree. The search starts at the root and explores
as far as possible along each branch before backtracking (Giarratano and Riley 2005).
In a forward chaining system, the process runs as follows: Initially, the inference engine searches
the conditional part of a rule, that is, the left-hand side of a rule, to determine if the input facts are
satisfied. The most popular rule control strategies to increase the efficiency of ES in terms of search-
ing through the rules in each cycle are the algorithms of Markov and Rete (Giarratano and Riley
2005). A rule whose conditional part is satisfied is said to be activated or instantiated and will be
put in a list called an agenda. If there is more than one activated rule, then they form a conflicting
set of rules. However, the system will only select one rule for execution or firing. The process that
the inference engine uses to give priority to a rule, when there are multiple rules that are activated
at once, is called the conflict resolution strategy. Available conflict resolution approaches have been
reviewed by Padhy (2005) and Jackson (1999). The result of firing a rule is the execution of the
then
part of the rule, that is, the right-hand side of the rule, which involves some actions or conclu-
sions; this may create new facts. This cycle is repeated until all of the possible information has
been extracted from the facts and the rules, and the system then reaches the final result. A similar
procedure with converse logic is followed for backward chaining.
Another popular and classic 2D AI representation technique is semantic nets (Giarratano and
Riley 2005). A semantic net uses propositional statements that are either true or false. In math-
ematical terms, a semantic net is called a directed graph that consists of nodes representing objects,
concepts or situations. Nodes are connected to each other by arcs called links which represent
relationships. Further to rules and semantic nets, frames have been also employed as a representa-
tion approach in many AI applications. Frames are very useful for representing casual knowledge
because the information is organised by cause and effect, which is in contrast to rules that are
based on unorganised causal knowledge. Frames are capable of representing either generic or spe-
cific knowledge about a narrow subject where some default knowledge already exists. Compared
to semantic nets, frames are able to handle a third dimension by allowing nodes to have structure.
11.5 BUILDING AN EXPERT SYSTEM
Several researchers have proposed various ES development methodologies, which are mainly based
on the classical waterfall model of the software life cycle (Buchanan et al. 1983; Medsker and
Liebowitz 1994; Turban 1995; Giarratano and Riley 2005; Padhy 2005). The linear model, which is
a type of life cycle model, has been used successfully in a number of ES projects (Giarratano and
Riley 2005). This type of model is usually employed in large commercial ES development projects.
For smaller, research-oriented prototype ES, not all the tasks or stages are necessary (Turban 1995;
Giarratano and Riley 2005). Thus, once a problem domain has been defined and analysed, the lin-
ear model can be condensed into the following three main stages: system design, development and
evaluation as described in the following text.
11.5.1 S
ySteM
d
eSign
This stage involves defining the system and then producing a detailed system design via the follow-
ing five tasks: system definition, knowledge acquisition, knowledge representation, knowledge base
building and the definition of the following - facts (or system inputs), control processes and system
outputs. In particular, system definition involves the statement of the aims and the scope of the sys-
tem. Knowledge acquisition is the process of extracting, structuring and organising knowledge from
