DEVELOPING KNOWLEDGE-BASED SYSTEMS: REORGANIZING THE SYSTEM DEVEPOPMENT LIFE CYCLE PRESENTER: SOPHIA FERGUSON.

DEVELOPING KNOWLEDGE-BASED SYSTEMS: DEVELOPING KNOWLEDGE-BASED SYSTEMS: REORGANIZING THE SYSTEM DEVEPOPMENT REORGANIZING THE SYSTEM DEVEPOPMENT

LIFE CYCLELIFE CYCLE

PRESENTER: SOPHIA FERGUSON

OBJECTIVESOBJECTIVES

1. EXPLORING THE ORIGIN OF THE KBS. Why build KBS?

2. UNDERSTANDING THE PROCESSES IN THE 3 STAGES OF THE KNOWLEDGE-BASED SYSTEM DEVELOPMENT LIFE CYCLE.

I. DEFINITION STAGE

II. DEVELOPMENT STAGE

III. INSTALLATION AND OPERATION

TRANSACTION PROCESSING SYSTEMS AND DECISION SUPPORT TRANSACTION PROCESSING SYSTEMS AND DECISION SUPPORT SYSTEMS SYSTEMS

VS.VS. KONWLEDGE-BASED-SYSTEMS KONWLEDGE-BASED-SYSTEMS

TPSTPS

• Transaction processing systems (TPS) perform routine data Transaction processing systems (TPS) perform routine data processing.processing.

• They are designed around forms, procedures, inputs and They are designed around forms, procedures, inputs and outputs, and often address well-structured problems.outputs, and often address well-structured problems.

• SDLC originated when most systems were TPSs.SDLC originated when most systems were TPSs.• facilitate, Royce’s or Boehm’s life cycle – Execute SDLC facilitate, Royce’s or Boehm’s life cycle – Execute SDLC

sequentially, with sign-off after each phase.sequentially, with sign-off after each phase.

TRANSACTION PROCESSING SYSTEMS AND DECISION TRANSACTION PROCESSING SYSTEMS AND DECISION SUPPORT SYSTEMS SUPPORT SYSTEMS

VS.VS. KONWLEDGE-BASED-SYSTEMS KONWLEDGE-BASED-SYSTEMS

DSSDSS• High levels on uncertainty.High levels on uncertainty.• A strategy to design first implement later is in-appropriate. A strategy to design first implement later is in-appropriate. • Users seldom know their requirements in advance.Users seldom know their requirements in advance.• Use heuristics and evolutionary strategies.Use heuristics and evolutionary strategies.

Prototyping Methodology

TRANSACTION PROCESSING SYSTEMS AND DECISION TRANSACTION PROCESSING SYSTEMS AND DECISION SUPPORT SYSTEMS SUPPORT SYSTEMS

VS.VS. KONWLEDGE-BASED-SYSTEMS KONWLEDGE-BASED-SYSTEMS

KBSKBS

• Depends on access to knowledge.Depends on access to knowledge.• High levels of uncertainty about project’s success.High levels of uncertainty about project’s success.• Applicable when problems cannot be solved with an Applicable when problems cannot be solved with an

algorithm.algorithm.• A nonlinear approach is applied.A nonlinear approach is applied.• Knowledge engineers must ferret out solutions.Knowledge engineers must ferret out solutions.

THE ORIGIN OF KBSDLCTHE ORIGIN OF KBSDLC

• Blue Cross/Blue Shield (BC/BS) and the Institute of Information Management, Technology and policy at the College of Business Administration, University of South Carolina formed a joint venture to build a knowledge-based system to automate medical review of claims. BC/BS selected an expert. The system was called MEDCLAIMMEDCLAIM.

BASIC SYSTEM-

BUILDING TASK IN

THE SDLC

THE KNOWLEDGE-BASE -SYSTEM THE KNOWLEDGE-BASE -SYSTEM DEVELOPMENT LIFE CYCLE DEVELOPMENT LIFE CYCLE

• The Knowledge-Based-System Development Life Cycle is a prototyping methodology for Knowledge-Based-Systems that uses expert system shells and programming environments.

PROTOTYPING METHODOLOGIESPROTOTYPING METHODOLOGIES

Although most KBSs cannot be pre-specified the Although most KBSs cannot be pre-specified the KBSDLC can be like --:KBSDLC can be like --:

• Evolutionary Prototyping by Incremental Development – where the final prototype is kept by the production system.

• Rapid prototyping – where the prototype is used as the design specification for a standard implementation of the production system, as in MEDCLAIM.

CONVENTIONAL METHODSCONVENTIONAL METHODS vs. vs.

PROTOTYPINGPROTOTYPING

• KBSDLC consists of KBSDLC consists of processesprocesses which are which are activated, deactivated, and activated, deactivated, and reactivated as neededreactivated as needed..

• Contrary to common Contrary to common practice with conventional practice with conventional SDLC, reactivation is SDLC, reactivation is desirable.desirable.

DEFINE PROBLEM DEFINE PROBLEM ANDAND

ASSESS FEASIBILITY ASSESS FEASIBILITY

• Definition of the business and knowledge problem is Definition of the business and knowledge problem is acquired.acquired.

• The business problem in MEDCLAIM was The business problem in MEDCLAIM was straightforward to define. The knowledge problem was straightforward to define. The knowledge problem was more difficult because the knowledge engineers were more difficult because the knowledge engineers were unfamiliar with medical review.unfamiliar with medical review.

FEASIBILITY STUDYFEASIBILITY STUDY

• A feasibility study is a quick examination of A feasibility study is a quick examination of the problems, goals and expected cost and the problems, goals and expected cost and benefits of the system.benefits of the system.

Feasibility StudyFeasibility StudyFeasibility StudyFeasibility Study

OrganizationalFeasibility

TechnicalFeasibility

EconomicFeasibility

OperationalFeasibility

FEASIBILITY STUDYFEASIBILITY STUDY

• Organizational FeasibilityIs the project aligned with the organization’s goal?

• Economic FeasibilityIs the project cost-effective?

FEASIBILITY STUDYFEASIBILITY STUDY

• Technical FeasibilityDoes the technology exist ?Does the technology exist ?What about the availability of staff to make the What about the availability of staff to make the

technology work?technology work?

• Operational FeasibilityWill the project improve the operation of the firm?Will the project improve the operation of the firm?Can the system be easily integrated into the current Can the system be easily integrated into the current

system?system?

IDENTIFY SUB-PROBLEMSIDENTIFY SUB-PROBLEMS

• Knowledge problems of any respectable size are too Knowledge problems of any respectable size are too difficult to comprehend as a whole, so the problem difficult to comprehend as a whole, so the problem should be broken down into workable sub-problems.should be broken down into workable sub-problems.

• Experience with MEDCLAIM suggests that Experience with MEDCLAIM suggests that identifying sub-problems is a significant enough identifying sub-problems is a significant enough activity to be a separate process.activity to be a separate process.

IDENTIFY AND DEFINE IDENTIFY AND DEFINE CONCEPTUAL STRUCTURECONCEPTUAL STRUCTURE

• A transaction processing system automates a process with A transaction processing system automates a process with many observable components, in contrast to a knowledge-many observable components, in contrast to a knowledge-based system, which automates a human problem-solving based system, which automates a human problem-solving process with few observable components.process with few observable components.

• Knowledge engineers search for concepts that characterize Knowledge engineers search for concepts that characterize the expert’s thinking about the problem.the expert’s thinking about the problem.

• Knowledge engineers should look at each concept (entity, Knowledge engineers should look at each concept (entity, attribute, or relationship) as equally important.attribute, or relationship) as equally important.

IDENTIFY AND DEFINE IDENTIFY AND DEFINE CONCEPTUAL STRUCTURECONCEPTUAL STRUCTURE

• In MEDCLAIM, providers and patients which in the TPS were key entities were peripherals details.

• Decisions were made about the claims as opposed to the patients.

• Critical concepts were:- The appropriateness of services rendered. The amount of services rendered.

CONCEPTUAL DESIGNCONCEPTUAL DESIGN

• Logical system design – serves the same purpose as it does in conventional SDLC.

• Creates a framework for using knowledge representation.

• Knowledge engineers select a knowledge representation (FOL, Semantic Network or Production Rules as in MEDCLAIM).

CONCEPTUAL DESIGNCONCEPTUAL DESIGN

• Knowledge engineers should be willing to make drastic changes and perhaps discard the prototype if needs be.

• To avoid information overload, knowledge engineers should activate new processes until a prototype is built to provide a focus point for departure.

DETAIL DESIGNDETAIL DESIGN

Knowledge engineers:

• Identify propositions for logics.

• Write descriptions and pseudo-code for procedures.

• Draw network diagram for semantic networks.

• Write English language rules for production rules.

• Draw diagrams or build models for direct representations.

• Identify and name slots for frames and scripts.

• Identify and name table entries for data.

CODECODE

• Translates detail design into the language of the knowledge engineering tool.

• Updating knowledge-base with knowledge information.

• Can run concurrently with detail design.

• Can create reactivations, as in MEDCLAIM.

TESTTEST

TEST REASONING SSearches for invalid reasoning.earches for invalid reasoning. To correct bugs coding is reactivated.To correct bugs coding is reactivated. Address mechanical details such as interfaces and internal Address mechanical details such as interfaces and internal

flow using artificial test cases.flow using artificial test cases.

TEST KNOWLEDGE

Correct code does not means correct knowledge.Correct code does not means correct knowledge. Attempts to detect invalid and ambiguous knowledge.Attempts to detect invalid and ambiguous knowledge. Invalid knowledge occurs when facts are stated incorrectly. Invalid knowledge occurs when facts are stated incorrectly.

VALIDATIONVALIDATION

• Detecting conditions missed earlier.Detecting conditions missed earlier.• Requires large sample of real cases.Requires large sample of real cases.• In MEDCLAIM, the reviewers ran claims through the In MEDCLAIM, the reviewers ran claims through the

system for several weeks and uncovered a few minor system for several weeks and uncovered a few minor problems.problems.

• Blind spots , due to hidden assumptions , therefore someone Blind spots , due to hidden assumptions , therefore someone else, with different hidden assumptions, should conduct else, with different hidden assumptions, should conduct validation.validation.

WHAT ‘S NEXT AFTER VALIDATIONWHAT ‘S NEXT AFTER VALIDATION??

• Use the prototype as the production system.

• Write off the prototype as a learning experience.

• Use the prototype as the specification for a conventional SDLC methodology.

WHO SHOULD BUILD THE KNOWLEDGE-BASED WHO SHOULD BUILD THE KNOWLEDGE-BASED SYSTEMS?SYSTEMS?

• Choose a expert who is Choose a expert who is articulate, cooperative and articulate, cooperative and willing to work with the willing to work with the knowledge engineer.knowledge engineer.

• Knowledge engineers need Knowledge engineers need a flexible approach to a flexible approach to problem solving, the ability problem solving, the ability to conceptualize and think to conceptualize and think abstractly and to listen abstractly and to listen without imposing their without imposing their views.views.

CONCLUSIONCONCLUSION

• A new SDLC philosophy is needed.A new SDLC philosophy is needed.

• Methodologies suitable for TPSs lack the adaptability Methodologies suitable for TPSs lack the adaptability needed for KBSs.needed for KBSs.

• Methodologies suitable for DSSs address the black box Methodologies suitable for DSSs address the black box problem but the knowledge engineer must open the black problem but the knowledge engineer must open the black box and get inside the expert’s head.box and get inside the expert’s head.

QUESTIONS???QUESTIONS???QUESTIONS???QUESTIONS???