Multi-Agent Architecture for Knowledge-Driven Decision Support - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 35

Figure 2. A schematic showing the generation, exchange, storage, retrieval and use of knowledge in
IKMDSA

Share
Domain
Knowledge

IKMDSA: l earning agent

IKMDSA:: Knowledge agent IKMDSA:: user agent

+Agent Knowledge : Domain Knowledge

+Learn() : Domain Knowledge
+ShareKnowledge() : Domain Knowledge

+ShareKnowledge() : Domain Knowledge +AcquireKnowledge()

+Retrieve *

*

Generates

domain knowledge

+Creates *

+Domain : Object = org.wc.DOM
+Rules : Object = org.WC.dom

ikMdsa user Interface generator
Welcome to the
IKMDSA User Interface
Please select a problem domain from the list
below:

* -Has Access to

+Store *

Raw Data
Repository

Tennis Anyone

OK

Knowledge Repository

User

site of knowledge storage. Intelligent agents deliver knowledge to the user interface to support intel-
ligent decision-making activity. The agent abstraction is built upon basic objects that take on additional
behaviors, as required by its function (Shoham, 1993). Knowledge exchange and delivery in IKMDSA
is facilitated through the exchange of the domain knowledge objects among intelligent agents. Figure
1 illustrates this basic building block of IKMDSA, where an agent has a composition relationship with
the domain knowledge object, and thereby has access to knowledge in the form of standard XML docu-
ment object model (DOM) objects.
Every agent can share its knowledge through the domain knowledge component by invoking its
ShareKnowledge behavior. The domain knowledge object contains behaviors to inform agents of the
name of the problem domain, share information about the various domain attributes that are pertinent
to the specific knowledge context, and share rules about making decisions for their specific problem
domain. These core components are used to develop the functionality of IKMDSA to learn rules and
domain attributes from raw data, create domain specific knowledge, share it with other agents and ap-
ply this knowledge in solving domain specific problems with a user. Once the attributes and domain
rules are captured in the domain knowledge object, using standard XML DOM format, they can be
exchanged between agents. Figure 2 provides a schematic of this activity sequence where knowledge is
created from raw data and ultimately delivered in usable form to the decision maker.

Learning Agents

Learning agents interact with a raw data repository and extract raw data used to generate domain
specific knowledge. Our model does not specify the storage representation and the data contained in
the repository may be of multiple representation formats including flat files, data stored as relational
tables that can be extracted using multiple queries into a recordset, or raw data represented using XML
documents. The learning agent extracts the raw data and applies machine learning algorithms to gener-