|
|
The KM is built on the basis of the following actions.
- Identify the task decisions made by the ES. The ES goal(s) should be already decided. Therefore, it is important to ensure that the goal(s) has/have been explicitly defined in such a manner as to be accepted by all parties.
- Design the goal decision block. The purpose of the ES is to make a decision about a concept. Place the concept in the center of a piece of paper. The concept will be one of the identified attributes. Refer to this concept as the goal decision. Frame this name in a rectangle.
- Add attributes to infer the goal decision. Place around the goal decision the attributes that will be used both in the formalisms (rules, frames, etc.) to infer the value of a goal decision and in the calculations to determine the goal decision.
Each attribute around the goal decision must be involved in inferring or calculating the value of the goal decision, but this does not mean that all attributes are used in a single formalism. Usually, there will be between three and six different attributes used in a number of rules, frames, etc., to infer the values of the goal decision. Most human brains can only handle four attributes at the same time. If over four attributes are used to infer the value of the goal decision, the expert will probably calculate the value of some intermediate attributes in order to infer the value of the goal decision. Generally, the number of attributes immediately around the goal decision can be reduced by introducing intermediate attributes.
- Extend the KM. Here, you need to add the attributes used to infer or calculate the values of each attribute employed to infer the goal decision. Continue extending the KM to include each attribute used to infer the value of another attribute. Gradually, a network of attributes can be built, resembling a "spider's web," though without interconnections between the edges.
Rules, and of course any other formalisms, may contain uncertainty, either because we are not absolutely sure how the rule has been represented or because the knowledge expressed in the formalisms is itself uncertain. In both cases, the link or relation between attributes is represented in the KM in the same way as relations without uncertainty. Rule uncertainty is recorded in the leaves of the rules.
- Enter multiple attribute uses. When many examples of an attribute are to be used in the left-hand side of a rule before the right-hand attribute is inferred, these examples are indicated with an M along the arrow connecting the two attributes. The expert may state when an example of a specific attribute value should be inferred in the decision-making process. When the links are not identified by M, they are assumed to be one-to-one, which is the most usual case. Apart from showing multiple examples of an attribute, the KM does not represent a data structure. In this sense, it is a representation of the process of inferring attribute values. If the KM is to be extended to include other attributes that can be used to infer the values of a multiple example of an attribute, this should be considered as a single example for purposes of KM extension. For example, whether other attributes can be used to infer the tool value required, rather than another that infers many values.
- Use the external states of the pseudoformalism, particularly pseudorules. Now we have to match the names of the attributes defined to the names of the concepts in the formalisms. The words used by the expert in the pseudoformalisms will probably have to be changed, taking care not to lose the meaning. The values of the attributes may be numerical, strings of characters or words, or one or many values selected from a particular list. The names of the attributes should be the same as those given to the expert. It is important to match the names and admissible values to the words in the pseudoformalisms, ensuring that they resemble natural language expressions as far as possible.
- Use a special notation to prevent duplicated inferences There may be hundreds of attributes in a large ES, many of which will be used in more than one place in the KM. Where the attribute value is inferred from one or more pseudoformalisms, it is not necessary to duplicate all the attributes used to infer the value. Instead, it is important to specify that the attribute has already been fully defined in another part of the KM. The KM box can be marked with an asterisk to identify duplicate attributes.
- Supplement the KM to include additional attributes and interrelations. Attributes continue to be added and the KM modified to reflect new attributes, values, and formalisms identified in the working document.
An example will better explain the manner in which a KM is built. In this case, we are going to design a KM for psychological diagnosis, confining the field of diagnosis to the different types of neurosis: anxiety, phobia, hysteria, and obsessive behavior, using the concepts obtained in preceding knowledge acquisition sessions: illness, patient, background (aspect of psychological interest in the patient's past), disorder, behavior, attitude, appearance, thought, fear. in Figure 9.
FIGURE 9 Knowledge map for the identification the type of neurosis.
The patient's attitudes, appearances, behavior, and background have to be analyzed to determine the type of illness. This is illustrated by the KM shown.
The expert may possibly know nothing about the patient's background, because he/she is reluctant to speak about his/her childhood. So, the expert must supply a rule or a default value to handle the situation. He/she may question the patient or deduce that if he/she does not wish to relive the past, it is because he/she does not want to go through it again. Additionally, the thoughts of the patient are difficult to analyze. Often, there are uncertainties, but the system must not stop. It must reason with the uncertainties. Reasoning may take longer, but the system has to make certain inferences from uncertain information, introducing other tests; for example, by putting further questions to the patient and checking them later.
|