Introduction
This section provides an explanation of the KADS methodology.
Objectives
By the end of the section you will be able to:
rdescribe KADS.
rUnderstand how KADS is an example of a PSM.
Purpose of KADS
In general information systems development, there are many methodologies that can be used to provide an overall control of that development.
Within KBS development however, there was no overall design strategy for a considerable time.
Activity 9
This activity will draw on your knowledge of information systems development to help you anticipate the need for something significantly different in relation to methodologies for KBS development.
Considering the process of developing an information system, what factors might make it difficult to directly apply such methodologies to the development of KBSs?
Feedback 9
You should have been able to recognise that most information systems are concerned with data and KBSs are concerned with knowledge. Data is gener- ally much more accessible—being stored in databases of one kind or another, whereas knowledge is stored in the minds of experts.
Knowledge-based systems also tend to be more complex. Information systems often perform numerical calculations which have a correct answer, e.g. a worker who works for 10 hours for £10 per hour should get paid £100. £99.99p while very close would be an incorrect answer. A KBS on the other hand simulates a human being making decisions. While decisions can be good decisions or bad decisions they cannot usually be clearly categorised as right or wrong.
For example, when choosing a university course, clearly some courses are in a subject that will interest the student more than others and some will lead to qualifications that will enable a graduate to find work more easily than others. Taking these factors into account you may choose to do a degree in computing. If you are not reasonably able in mathematics doing an engineering degree may be a poor choice for you—but this decision could not be clearly categorised as a wrong decision. Thus checking the quality of the outputs from a KBS is much more difficult than checking the outputs from an information system.
The KADS is an attempt to overcome this difficulty by providing a system for knowledge engineers and ES developers to follow.
Knowledge acquisition design system aims to solve two specific problems in KBS development:
r Firstly, large-scale problems could not easily be solved by one knowledge base—
especially if it was restricted to one knowledge representation scheme—and which was very inefficient to run and difficult to maintain. This was overcome by the development of blackboard architectures and the same principle of segmented knowledge bases is supported by KADS.
r Secondly, the benefits of explicitly separating control and the domain knowl- edge became clear as the modelling approach was adopted, and thus the KADS methodology was developed as a problem-solving methodology (see previous section).
Knowledge acquisition design system, and its more recent variant, Com- monKADS, is the most commonly used methodology within Europe for the devel- opment of KBSs. KADS is the most prominent example of a PSM-based method- ology, thus discussions in the previous section apply to KADS.
Knowledge Acquisition in KADS
The KADS approach includes the following knowledge acquisition activities:
rElicitation—eliciting the knowledge rAnalysis—interpreting the knowledge
rFormalisation—formalising the knowledge so that it can be used in a computer.
Before KADS, there was an approach to knowledge acquisition that consisted simply in:
racquiring domain knowledge
rtransferring knowledge (somehow) to a KBS.
In this approach the experts reasoning process was not modelled—it was left to the inference engine to determine if/when to apply the knowledge.
The KADS approach treats the knowledge-acquisition process as a modelling activity, i.e., the expert’s problem-solving knowledge is modelled, among other models, this leads to the efficient application of domain knowledge and allows reuse of control and domain knowledge.
Multiple Models in KADS
Based on the ideas of modelling the PSMs, KADS supports the development of various models. These include:
Process or organisation model, where the processes within the organisation are modelled in order to assess the role and impact of the KBS. This reduces the fric- tion that may occur when trying to implement the KBS within the organisation.
Expertise model, models the problem solving or expert behaviour required of the system. Knowledge acquisition design system libraries of reusable PSMs have been created to support prediction, assessment, design, planning and schedule tasks.
Activity 10
This activity involves you in discovering the characteristics of some of the other models used in KADS.
Other models that may be used in the KADS approach include:
rApplication model—defines the functions of the system with respect to users rTask model—defines the tasks that the KBS must perform
rCooperation model rConceptual model rDesign model.
1. Search the Internet for documents relating to the last three of these models.
2. Make brief notes on their purpose.
Feedback 10
You should have been able to locate documents describing the models as fol- lows:
Cooperation model—specifies how subtasks in the task model should be done if cooperation is necessary. This model would need to be applied if, for example, the solution of a problem by the system required information from the user.
Conceptual model—this is essentially a combination of the models of expertise and cooperation as these together specify the overall behaviour of the system.
Such a model would be based on abstract descriptions of the objects and oper- ations that the system needs to know about.
Design model—specifies how to implement the system in the form of descrip- tions of computational and representational techniques as well as hardware and software requirements.
Theses models essentially represent steps in defining the goals of the KBS devel- opment.
Some of the advantages of this multiple model approach in KADS are that those involved in the development of the KBS can more easily identify, describe and select characteristics of the targeted system as well as focus on specific aspects while ignoring—at least for the moment—other components.
Activity 11
This activity will help you visualise the relationship between the various models used in the KADS methodology.
Complete the following diagram that illustrates the relationship between the various models.
Organisational Model ____________ Model
Task Model
______________ Model __________ Model
Conceptual Model
__________ Model System
Feedback 11
Your completed diagram should look like this:
Organisational Model Application Model
Task Model
Cooperation Model Expertise Model
Conceptual Model
Design Model System
The KADS Four-Layer Model
As well as using modelling as a technique for describing the components of a KBS during the process of its development, KADS views a completed KBS using a model with four layers. This four-layer model provides a method of representing knowledge within an ES (see Figure 6.6).
Domain Layer
The domain is the static knowledge in the KBS. The basic knowledge and some relationships are recorded by the knowledge engineer using the linguistic level of the conceptual model.
Domain Inference
Task Strategy
FIGURE6.6. The four-layer KADS model.
The Inference Layer
Knowledge is grouped into related units, or meta-classes and a useful classifica- tion system is devised. Relationships between knowledge may be identified and recorded in frames or semantic networks. In this layer, knowledge is being trans- formed from the linguistic level to the conceptual level.
Task Layer
This layer describes how the domain knowledge and inferences from that knowl- edge can be used to solve a specific task. It uses the conceptual links in the knowl- edge and then attempts to add the epistemological relationships.
Strategy Layer
This layer deals with the overall approach and planning involved in solving a problem. The aim is to identify problems in the knowledge-building process (e.g.
inconsistent rules) at an early stage in the system design process. It therefore at- tempts to place a formal structure on knowledge, moving from the epistemological level to the logical level prior to implementation.
Activity 12
You are a knowledge engineer attempting to elicit knowledge to build a new ES. The knowledge domain is weather prediction. You are currently working on a module to forecast the amount of rainfall.
Knowledge from the expert indicates that there are many variables affecting accurate weather forecasting including wind speed and direction as well as overall temperature, not only on the ground but also in the clouds. The expert notes that rainfall may be preceded by a fall in temperature, while winds blowing off the sea to the west provide an increased chance of rainfall.
As far as the information allows, start to produce a four-layer model for a KBS by outlining which components of the model will refer to the information and knowledge available.
Feedback 12
You should have been able to produce an outline similar to the following:
Domain layer
Data concepts include: temperature, wind speed, wind direction, amount of cloud cover, etc.
Inference layer
Inferences from the data concepts will include:
Falling temperature indicates increased probability of rain.
A westerly wind from the sea normally provides increased chance rain.
Task layer
The basic task or reasoning technique is chosen, with a decision being made concerning task-driven or goal-driven structures (or alternatively a choice will be made between backward or forward chaining).
Strategy layer
The formal structure of the knowledge is identified (this is a forecasting model) and the appropriate data structures designed.
Strengths and Weaknesses of KADS
A major advantage of the KADS approach is in the idea of generic task models (GTMs), also known as interpretation models. These can be thought of as skele- ton models for typical tasks or task fragments, such as ‘classification’ or ‘system diagnosis’ stored in generic task libraries. Knowledge engineers can use suit- ably chosen GTMs to guide the knowledge-acquisition process in a new domain, refining and combining GTMs to produce a fully specified model.
KADS does have weaknesses, for example:
rIt is difficult to translate between or connect the different layers.
rAll the layers are rarely used; most people tend to use the diagrams, but these are not expressive enough for all requirements.
rKADS systems typically end up with large amounts of documentation for rela- tively modest systems and are hard to change.
rKADS does not itself specify the representation types to be used in describing its various models.
The last point is important, since we must decide what our needs are for repre- sentations. Suitable representations must have a two-sided functionality, i.e., they must be able to:
rexpress the language of the testing techniques
rdescribe systems in such a way that they are recognisable to those who must contribute to the development of evaluation models.
Summary
In this section you have seen how the KADS methodology can be used to construct a variety of models to bridge the gap between required behaviour and system behaviour for a KBS.
Web Links
CommonKADS
http://www.commonkads.uva.nl/frameset-commonkads.html