| Background Expert systems have traditionally been
developed with most of the knowledge buried in the program logic.
Programmers were needed to maintain the expert rules.
This approach made developing the rules slow, expensive and error
prone. The key to a successful expert system is a flexible database
that allows the experts to
maintain the rules directly. The expert logic must be encapsulated in
the database and not in the program logic. The interface used to
transfer the knowledge from the expert to the database must be
robust, easy to use, flexible and allow for easy validation..
Can Logic Be Fuzzy?"Fuzzy Logic" is a term
dealing with approximate reasoning rather than precise reasoning.
Our data-driven expert system analyzes situations. Situations are
described by the characteristics that describe the system both in a
positive and negative manner. Expert should be able to model the situations that they are dealing
with to see how the knowledgebase deals with it.
This does not say that an expert system always yields an exact
outcome. The knowledge stored in an expert system has limits. The
limits are in the scope of the system and the depth of it's
knowledge. Building an knowledge base is not a project with
a fixed completion point. It's a ongoing process, The information
and knowledge that we continually acquire needs to fed to the expert system.
However, since the knowledge is in the database not the program
code, the program that maintains the database can be completed.
Expert Logic vs. Sequential Logic
Programming logic has traditionally been sequential. The logic
flows sequentially and branches based on predefined
conditions. Expert logic is situational. Every situation is
different. The program needs to respond to the information that is
being gathered from the user. The basic logic flow in an expert
system is:
- Capture information
- Evaluate the information
- Draw a conclusion.
- Request more information
- Act on the information
- Additionally, the user may be able to query the system to
determine the possible solutions and outcomes in the midst of a
situation. This is useful for the expert when they are
developing the rules and for the user when the expert system has
reached the limits of it's knowledge.
What Can an Data-Driven Expert System Do?
Data-Driven Expert System excel at investigative work. An
investigative system gathers data. The data is analyzed to determine
what additional data needs to be gathered to complete the
investigation and draw the proper conclusion.
Typical acts performed by the expert system depend on the
application:
- Investigative or Audit
- Send an e-mail, letter or fax
- Initiate a phone call
- Request a credit report or other background document
- Initiate a transaction with another system.
- Determine whether the case is satisfactory or defective
- Score an application.
- Help to uncover fraud.
- Reveal practices contrary to regulatory or corporate
guidelines.
| - Medical
- Diagnosis
- Recommend dosage levels
- Determine admission status
- Profile risk
- Consumer Goods and Services
- Troubleshooting
- Automated support
- Determine admission status
- Profile risk
- Telemarketing
|
Benefits of a Data-Driven Expert System
- Reduced training requirements. The computer is trained to do
the expert analysis. Less skilled people can operate the system
and interpret the results.
- Consistent Analysis.
- Overall effectiveness.
- Tailored responses to individual situations.
- A readily available source of information.
- Speed of Analysis.
Keys to a Successful Data-Driven Expert System
- Fast. Situations must be analyzed as fast as the user can
enter information.
- Easy. Entry must be easy. Voice recognition systems will be a
key factor in many applications.
- Powerful User Interface for Rules input and analysis.
- Ability to handle large amounts of data
- Easy access to reference materials.
- Statistical analysis capabilities.
Example Expert System Data Gathering Screen
This screen illustrates one way to capture information. The user
categorizes information into sections. Some sections can occur once
in the case, others multiple. Each section can be completed
independently. The Questions for each section are shown on the lest
with their answer. The highlighted question has additional
information and reference material available on the bottom. The
bottom along the right allow the user to perform special functions
like collect special information, or manage the actions created for
the situation.
The screen used for mortagage auditing by many leading lenders.
Construction of the Rules Database
The internal design of the rules database has many components
- Questions of many different types need to asked. The various
formats need to be handled.
- The possible answers value that have an affect recording.
- The facts need to be grouped into conditions. When a condition
is met, several things could happen.
- A series of questions could be asked or actions staged for
execution.
- A condition could prevent an action or question from being
asked.
- A conclusion could be reached. It could need further support
and documentation for reporting.
Mitten Software Can Help Develop Your Expert System.
Contact 952-745-4941, ask for Jim.
Or send an e-mail
to answers@mittensoftware.com
|
