|
|
- 3. Organize case-independent facts. So far, we have discussed that an ES reasons with facts and hypotheses about case attributes. However, most ES also use facts about application-dependent areas that are independent of any specific case. Just as a concept type serves to organize attributes, the concepts themselves may serve to organize the case-independent facts. This involves: (1) revising the notes on knowledge acquisition sessions and identifying the case-independent facts that specify the values of concept attributes, considering only the facts to be used by the ES when executing a task; (2) identifying the concept -- process, object or entity, etc. -- to which each fact belongs; (3) grouping the facts and groups of facts by the concepts to which they belong; (4) listing the attributes that specify the facts about each concept type; and (5) developing intermediate representations that summarize collections of related facts.
So, each case-independent fact specifies the value of an attribute of a particular concept. These facts take the form: "a general concept attribute is value," for example, the patient's temperature is 39C.
- 4. Identify case-independent relations. Many ES use case-independent facts to express relations between different concepts. A relation may be expressed as "concept 1 relation concept 2"; for example, cable x is connected to plug y. Alternatively, a relation may take the form: "relation between concept 1 and concept 2 is value," for example, the distance between the computer and the wall is 15 cm. The notes on knowledge acquisition have to be revised to identify relations that are not facts and that will be used by the ES in the reasoning process. Here, we have to (1) identify the facts specifying relations between concepts, looking particularly for expert statements concerning taxonomies, structures and other types of relations; (2) group facts specifying the same type of relation; (3) check each relation against the strategic and tactical knowledge gathered, separating the relations used by the ES from additional relations providing an understanding of the basic reason for system actions and checking whether the relations have already been captured in the tactical knowledge representations; and (4) develop intermediate representations to illustrate the relations to be used by the ES that have yet to be captured in the intermediate representations of tactical knowledge.
The most useful representations for relations are graphs and trees, in which the nodes correspond to concepts and the arcs to relations. Generally, each graph would illustrate a simple relation type; for example, part of or cause. Arcs are used to indicate the direction of an asymmetric relation.
3.4. CONSTRUCTION OF THE STATIC SUBMODEL: CONCEPT DICTIONARY
Having identified the strategic, tactical, and factual knowledge during the analysis stage of the conceptualization phase, we move on to synthesis, which begins with the construction of the concept dictionary.
TABLE 4 |
Concept Dictionary for any Disease |
|
Concept |
Function |
Synonyms/acronyms |
Attributes |
Inferred from |
|
Clinical record |
Record patient treatment and clinical record |
Case history, CH |
Name, age, sex, date of treatment, test, results, allergies, etc. |
Tests and care |
Medicines |
Cure disease or relieve symptoms |
Treatments, prescription, chemical formula |
Chemical group, proprietary name, active ingredient, contraindications, dosage |
Treatments, texts, vade mecum, etc. |
Diagnosis |
Treatment prescription |
Examination, tests |
Causal agent, side effects, differential diagnosis |
Clinical symptoms, signs, test results |
Treatment |
Cure the patient |
Patient management |
System, mode of administration |
Diagnosis, texts, physician experience |
|
The identification of high-level concepts or functional subconcepts and key terminology is described in a concept dictionary. For each concept, the dictionary will specify its utility or function, synonyms and acronyms, defining attributes, values, data source, etc. It is important to mention that dictionary construction is not a quick job, and that it will not be amended later. Indeed, it is a living list, far from easy to construct, which synthesizes and summarizes everything obtained using the different knowledge acquisition methods. Table 4 shows part of the concept dictionary of a medical diagnostic and treatment system. Any other inputs deemed appropriate, such as cross-references, may be added. The important point about this simple example is that it identifies the components essential in any dictionary and how they should be specified.
The concept dictionary is useful because it isolates prominent particulars, that is, attributes, properties, or characteristics that distinguish the task concepts. Similarities and differences are explained and explicitly recorded. Questions, such as what makes A, B, and C similar?, what is the difference between A and B?, or is there any overlap or common point between A and B?, may be asked in order to identify these similarities and differences. When checking for similarities or differences, which may not always be immediately obvious, a possible technique is concept classification, whereby a list of concepts, taken of course from the concept dictionary, is given to the expert, and he/she is asked to order them in two sets. Each set is given a name or label that best describes why they were grouped in this manner. The process is repeated with three sets and so on. The concepts related with common properties or particulars are grouped and represent a concept classificational taxonomy or tree. The classificational and distinguishing properties are specified at the bottom of the tree as it is built. The classificational tree is one of the most natural means of representation and is also easy to understand, proving to be a very useful technique for using organizational knowledge and representing hierarchical classifications, where the lowest items in the hierarchy inherit the properties of their ancestors.
This concept identification and recording stage is a prerequisite for later conceptualization steps. It also enables the KE to understand and define the meaning of certain important words most commonly used by the expert. A possible method of identifying these concepts and building the associated concept/attribute/value table is given below.
- Underline, for example in red, each name or noun mentioned by the expert.
- List all the underlined names sequentially, as they appear in the working document, to form a Glossary of Terms.
- Examine the full list and group as many names as possible on a new sheet to form classes of things, that is, concepts. In other words, a concept may be unique, a class, or a related group of concepts. Concepts may be, as discussed earlier, objects, items, or events with attributes and values. The concepts and their interrelations reflect how the expert views or interprets the problem domain. By comparing the names on the original list with the list of concepts, the grouping of many will be immediately obvious from the structure of the document or interview.
|