Information Technology Reference
In-Depth Information
the choices made by the agent. This is achieved through parsing the decision rules based on the param-
eters supplied by the user. The agent compares the users' selections with the known rules and decides
on the rule(s) that are fired for the given instance. These rules are formatted in a user-friendly format
and made available to the user. This provides the user with a decision, given their selection of domain
attributes and provides the user with explanations of the decisions made, given the selections made by
the users.
The above sections provide a complete description of the process of knowledge creation, knowledge
representation, knowledge exchange, KM and the use of the knowledge for decision making employed
by IKMDSA. Figure 3 provides a schematic of this overall process. As shown in figure 3, IKMDSA is
designed for a distributed platform where the knowledge available to the agents in the system can be
made available on an intranet and an Internet based platform by enclosing the domain knowledge objects
in SOAP wrappers that enables the knowledge broker functions of the knowledge agent by making its
knowledge available as a Web service.
IKMDSA consists of intelligent agents as discussed above that are able to provide intelligent deci-
sion support to the end-users. All of the agents in the architecture are FIPA compliant in terms of their
requirements and behavior. The learning agents create knowledge from the raw data in a data repository,
knowledge agents primarily acquire this knowledge from learning agents and manage this knowledge
through a knowledge repository, while user agents help the users make decisions on specific problems
using the knowledge contained in the decision trees. The exchange of knowledge between agents and
between users and agents is achieved through sharing of content information using XML. The agents
work on a distributed platform and enable the transfer of knowledge by exposing their public methods
as Web Services using SOAP and XML. The rule-based modular knowledge can be used and shared by
agents. Capturing the modular knowledge in XML format also facilitates their storage in a knowledge
repository - a repository that enables storage and retrieval of XML documents. The architecture allows
for multiple knowledge repositories depending upon the problem domain. The benefits of such knowledge
repositories are the historical capture of knowledge modules that are then shared among agents in the
agent community. This minimizes the learning curve of newly created agents who are instantiated with
the current knowledge that is available to the entire system. This is achieved in IKMDSA since agents
have captured rule-based knowledge modules and have stored such knowledge modules in XML format
in the knowledge repository for the benefit of the entire agent community and the system.
IKMDSA also provides a decision explanation facility to the end-users where agents are able to
explain how they arrived at a particular decision. This has three important benefits:
i. Tthe end-user can understand how the decision was made by the software agent
ii. The end-user can make a clear assessment of the viability of the decision
iii. The end-user can learn about the problem domain by studying the decision paths used by the
agent
Agents are able to explain the rules and parameters that were used by the agent in arriving at the
stated decision. This explanation facility is a natural extension of using decision trees in general for
solving rule-based decision problems. Non-technical end-users are able to easily understand how a
problem was solved using decision trees compared to other existing problem-solving methods such as
neural networks, statistical and fuzzy logic-based systems (Sung et al., 1999). The IKMDSA architecture
can provide intelligent distributed decision support that may be internal to the company and the other
Search WWH ::
Custom Search