Design of Structured Knowledge Models

The question answering capability of intelligent interfaces according to the FLUIDS conception resides in a general framework of structured knowledge for real-time decision support that can be instantiated for any concrete application domain. This framework identifies the different knowledge elements occuring in a generic decision making scenario, and associates them with their corresponding roles in each step of the decision support process. Moreover, the generic model includes knowledge to associate potential queries of a human decision maker with appropriate problem solving methods which are needed to derive suitable answers to these questions.

The FLUIDS methodology for building up a declarative model in terms of knowledge structures relies on the notion of knowledge unit as a fundamental block for knowledge modeling. A knowledge unit encapsulates specific knowledge structures (so-called knowledge areas ) with their associated reasoning capabilities (so-called tasks ). This idea is exemplified in Fig. 1.

 

Figure 1: An Example of a Knowledge Unit 

Every single knowledge area can be represented in terms of a subsequent knowledge unit which leads to a decomposition into a knowledge unit hierachy. The leaves of this hierarchy are called primary knowledge units. In order to obtain a specific knowledge architecture for a given domain, the primary knowledge units are associated with suitable representation primitives from a library of reusable knowledge representation components. Common domain concepts that are to be shared among different primary knowledge units are grouped into conceptual vocabularies.

During the knowledge acquisition step for a specific application domain, the generic knowledge model is applied and further specialized in order to obtain a concrete instantiation of the knowledge architecture for the given domain.

 

Figure 2: FLUIDS Knowledge Model for Providing Decision Support 

Fig. 2 represents FLUIDS ' generic structured knowledge model for real-time decision support. In this model, two different knowledge layers can be distinguished:

• Conversation management is concerned with the dynamic association of questions formulated by the user to appropiate problem solving methods.
• Decison support problem solving provides the necessary problem solving capabilities needed to obtain suitable answers for a given query.

The problem solving layer includes knowledge structures related to the different facets of the abstract decision support problem:

• Identification knowledge relates the states of the involved system components to specific problem situations.
• Diagnosis knowledge makes it possible to find the causes for the identified problems.
• Regulation knowledge is needed to propose corrective control actions depending on the problem situation and the corresponding diagnosis.
• Prediction knowledge offers support for the determination of the possible evolutions of the system given its current situation.
• Coordination knowledge allows the integration of control actions to be performed over different system components in order to ensure a globaly consistent regulation proposal.

Such a generic problem solving scheme can support a variety of potential query patterns to be handled by conversation management layer. The following patterns characterize the basic question types required for decision support:

• ``What happens ?'' at the basic as well as the global level;
• ``What may happen ?'' and ``What may happen if ...?'' ;
• ``What to do ?'' and ``What to do if ...?'' .