![]() |
|||
![]()
|
![]() |
![]() |
![]() |
3.7. VERIFICATIONSHaving designed the two submodels (static and dynamic) and integrated them into the KM in the synthesis stage to build the conceptual model, a series of checks have to be carried out to eliminate subjectivities, consider unknown conditions and/or address uncertainty, and verify the completeness and consistency of the whole model. All these questions will be considered in more detail below. Verify responses to eliminate subjectivities. Having designed the preliminary KM, consider the peripheral attribute values. There are only three ways of establishing the values of the above attributes, namely: obtaining the value from an external source, i.e., a file; asking the user to enter the value; or using a default value if neither of the above methods can be used. As it is not possible to infer the values from an intermediate representation in the case of peripheral attributes, it is important to examine the possible values assigned to these attributes and identify the values that would force the user to make a subjective decision. Moreover, intermediate representations should ignore user subjectivity. Any subjectivity should be contained in the expert rules. The KE should not ask questions or raise issues that imply evaluations. All issues and questions should be related to the facts. Examine unknown and default conditions. It is also important to ensure that the KM and the dynamic submodel, especially the pseudorules, contain links and values to handle unknown conditions. If there is a possibility of the user not knowing an attribute value, it is important to ensure that the expert supplies either a pseudorule or any other pseudoformalism to handle the situation or an acceptable default value. If no pseudorule or default value exists, unknown conditions may arise. This problem must be added to the list of questions in the next structured interview with the expert during the next knowledge acquisition session. Verify negative conditions. The pseudoformalisms not only should state what to do when given conditions are met or arise, but also provide a means of expressing what to do when such conditions do not exist, the expert often knows what to do when the expected conditions do not occur or exist. Therefore, it is important to check with the expert during the knowledge acquisition whether the non-existent conditions are important and whether there are potential ways of covering such cases. Address uncertainty. Any pseudoformalism or information may contain uncertainty. It is important to confront experts about what to do if a fact is uncertain. Additional pseudoformalisms that incorporate uncertainty may be needed to address these cases. There are few ES that use certainty factors in the formalism. Certainty factors are not the same as probabilities. If they are required, procedures must be defined to manage them. Verify completeness and consistency. Check that each known attribute value is used in a knowledge representation pseudoformalism. All the values should be in the Concept/Attribute/Value table. Unless the attribute is in the periphery of the KM, the KE must ensure that each value can be inferred from a pseudoformalism or obtained from a default value. If the values are not used, ask the expert why. Also, it is important to verify that each attribute value, except the goal decision, is used in the left-hand side of a rule or appears in any other place in association with an attribute. Otherwise, ask the expert why it has been entered as a possible value and has not been used. Additionally, it is important to verify the KM to ensure that a value does not follow a recursive loop in which the attribute is inferred from the second value of a second attribute, which is, in turn, inferred from a value of the first attribute. Similarly, unnecessary IF conditions, subfunctions, etc., must also be verified. 3.8. SYSTEM DECOMPOSITIONWhen the static and dynamic models have been completed, it is often possible and desirable to decompose and segment the ES. Sometimes, segmentation is left until several ES development process cycles are completed and the KM illustrates system complexity. Segmentation is possible if users accept subsets of the ES until the full system is completed. It is important to underline that this segmentation is not incompatible with the decomposition of a task into subtasks. On the contrary, whereas the decomposition is carried out during knowledge analysis, segmentation is carried out when the conceptual model is almost finished. Additionally, while decomposition into subtasks strictly concerns the conceptualization phase, segmentation also involves the implementation phase. Segmentation is desirable if it reduces development time and maintenance effort. Segmentation will only produce this reduction if there are many concepts, attributes, and values proper to each system subset. If a subset uses over 30 different attributes and over 100 different attribute values, times will be reduced significantly if the subset is developed separately. Segmentation is carried out in the following stages.
The KM can be used to identify the subsets of the segmentation. In this case, it is important to look for attributes that repeatedly appear in the KM.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |