SOME
ASPECTS REGARDING THE INTEGRATION
OF INFERENCE PROCESSES IN KADS SYSTEMS
Dorina Petrică
Department of Automation and Industrial Informatics
"Politehnica" University of Timişoara
e-mail: dpetrica@aut.utt.ro
Abstract. The KADS (Knowledge
Acquisition and Design Systems) method has known a continuous development due
to its logical (natural) structure of approaching knowledge based systems. On
the one hand, the paper illustrates the way in which KADS methodology may be
applied to the expert systems, on the other hand the way in which the
qualitative table of inference (QTI) structure used to describe in detail the
inferential processes may become a KADS tool. Within this context a minimum
expert system for isolated paralysis of extraocular muscles is used as an
example.
INTRODUCTION
A successful development of a project in informatics requires a methodology which should offer those who work within the project, a common language that facilitates communication among decision makers, users, designers and executioners. The importance of the method is much emphasized within knowledge based systems, as the achievement of such systems usually requires the collaboration of several multidisciplinary teams. The method is in fact a guide for the designers, a guide in which one can find a minute description of the development stages of a knowledge based system, from the identification of the problem to the final solutions evaluation. The methodological background assures the quality of the solution, the control of the development process and the quality of the final product. KADS (Knowledge Acquisition and Design Systems), itself, is such a cognitive modelling oriented method, created within the ESPRIT [1] project.
The development cycle of an expert system proposed by KADS is structured in the following steps : 1- initial study (defining the application) ; 2. analysis (specifying the application) ; 3. conception (modelling the application) ; 4. implementation (programming of application) ; 5. testing (validation) ; 6. installation (loading of knowledge bases) ; 7. use (interrogating the knowledge bases) ; 8. maintenance ( actualizing, adapting, saving, restoring). This way of structuring the whole process in steps is found in KADS methodology in a hierarchy of models which succesively transform themselves into one another; and thus the development process is directed from the user’s specification to the real informatics system, implemented and set up.
In the analysis stage the problem to be
solved is examined from three different points of views: the restrictions to be
respected, the expertise tasks to which it responds, the usage scenarios. The
analysis together with the conception step should assure both knowledge
formalizing, and its processing. This involves a full understanding of the
studied field, which understanding can be obtained by the construction of the
following models associated to the expertise: conceptual model, functional
model, logical model and physical model.
EXPERTISE MODELLING
Approaching the process of knowledge acquisition by modelling, means a structural approach which assures the access at the cognitive level, by moving the expert system designer’s effort from the formalizing area to the modelling area. The modelling differentiates the domain description from the reasoning description, as the constituents of the global model are structured on four levels.
- domain level (consists of the description of objects, concepts, attributes and relations);
- inference level (consists of the descriptions of reasoning primitives) ;
- activity level (consists of the description of reasoning stages, sequences of activities and control elements ;
- strategy level (consists of the general plan of problem solving).
These levels along with the expertise may be found in the structure of the reasoning model for diagnosis problems [2].
Taking into account this
reasoning model, a rule based expert system was designed to diagnose the isolate paralysis of
extraocular muscles, on the basis of the following clinical semiology doctor –
patient elements analysis; the defective position of the head and neck
(step I), the repose position of the eyeballs (step II), the directions of the
limitation of eyeball movement (step III), diplopia in its dynamic (step IV). The
concepts and the expertise strategy were established as a result of a knowledge
acquisition process schematically represented in figure 1, which emphasized also the place of the
synthetized models in the designing process.
Fig.
1: Synthetized Models by knowledge engineer
The conceptual model associated to the four steps representing the examination procedures led to the identification of 60 concepts (F1..F60), respectively, the logical models associated to the steps led to the identification of 110 fixed rules, which can be reduced to 44 variable rules (R1...R44).
The conceptual model associated to the
expertise allowed the emphasizing of two diagnosis levels : level 1, which
corresponds to a conclusive diagnosis after execution of the steps I and II, so
the diagnosis can be further checked with step IV; level 2, which corresponds to the situation in which the
execution of steps I and II doesn’t lead to a conclusive diagnosis, in which
case the execution of step III is necessary, compulsorily followed by step IV.
The logical model associated to the expertise is represented by a knowledge
base which consists of rules, obtained by a description in detail of some
supplementary activities (conclusive diagnosis test,
decision for checking, etc.) and by naming an inference strategy. The logical
models, both at the step level and
the expertise level have been analyzed by using the qualitative table of
inference structure.
QUALITATIVE TABLE OF INFERENCE (QTI)
The QTI represents a scheme which illustrates
the way the facts
and rules, associated to a knowledge base interact. Taking into account for the
inferential process at the steps level the backward chaining strategy and the
sequence < step I, step II, step III, step IV> as a sequence for the
examination procedures, the knowledge base RE-CP1… RE-CP 10 resulted, knowledge
base which consists of rules in the Horn clausal form.
RE-CP1. If "Start"
then "hypothetic
diagnosis"
RE-CP2. If "hypothetic diagnosis"
then "execute-Step
I-CP"
RE-CP3. If "execute-Step I-CP" = =
confirmed
then "Diagnosis-confirmed"
RE-CP4. If "execute-Step I-CP" = =
invalidated
then "execute-Step
II-CP"
RE-CP5. If "execute-Step II-CP"= =
confirmed
then "Diagnosis-confirmed
"
RE-CP6. If
"execute-Step II-CP"
= = invalidated
then "execute-Step III-CP"
RE-CP7. If "execute-Step III-CP" = =
confirmed
then "Diagnosis-confirmed "
RE-CP8. If "execute-Step III-CP" = =
invalidated
then "execute-Step IV-CP "
RE-CP9. If "execute-Step IV-CP" = =
confirmed
then "Diagnosis-confirmed "
RE-CP10 If "execute-Step IV-CP" = =
invalidated "
then "Diagnosis-invalidated "
The QTI associated to this knowledge base has the configuration of figure 2. The rules are represented by rectangles and the facts by rounded rectangles. The convergence of several rules to the same fact has the significance of the connective OR applied to the rules by which that fact can be deduced. The connective AND applied to the facts of the rule premises (situation which is not illustrated in figure 2), is represented in QTI by the convergence of several facts to the same side of a rule. QTI makes a hierarchy of the rules from the point of view of the order in which they can be activated and, correspondingly, a hierarchy of facts to be used in order to point the reasoning within the state space. If we look at the table “bottom – up”, we can find out the facts definable as goals. If we look at the table “top – down” we can find out the necessary facts to demonstrate the goals. So, this chart provides useful information for both basic inference strategies, which fact has been suggested by the arrows P→C (premises → conclusions), respectively C→P (conclusions → premises). Meanwhile, QTI allows a visualization, in a compact form, of all elementary inferences possible to be made in the inference processes associated to a statement.
Fig. 2: QTI associated to the expertise
If we consider the QTI in the direction P→ C we can see that it corresponds to the mixed inference strategy, the rules RE-CP1...RE-CP10 being submitted to an inference process by “forward chaining”, and R1..R44 (steps level) by "backwards chaining". Considering the QTI in the direction C→P, with the goal "Confirmed diagnosis" coresponds to the backwards chaining strategy. The order in which the examination procedures are executed may be controlled by the placement order of the rules in the rule base R1...R44.
CONCLUSIONS
Approaching the knowledge acquisition process by modelling assures a systematic standardization of both concepts and strategic and control knowledge. Conceptual and logical models assure the necessary elements for achieving, with less effort, the functional and physical models associated to the implementation of expert systems. QTI is a useful instrument for KADS method as well, for the clarification, simplification and reduction of reasoning process associated to any level.
References
[1] Brunett E.: KADS:
Methode d'ingenierie de la connaisance, Genie Logiciel &
Systemes Experts, nr. 23, June, 1991.
[2] Petrică, Dorina: Contribuţii la aprofundarea mecanismelor inferenţiale ale sistemelor expert cu aplicaţii în domeniul medical, Teză de doctorat, 2000.