Copyright(C) Paul C Caswell 1996-2008 All Rights Reserved

Navigation Menu
General
Home
Contact Us
 
Standard ESB
Overview
Question Editor
Knowledge Acquistion
User Interface
Screenshots
Donate with PayPal

 

Internet ESB
Overview
JavaESB v1
JavaESB v2
 
Getting Help
Discussion Forum
Online Tutorial
FAQ
 
Buy Online
JavaESB v2
 
Free Downloads
ESB (Standard)
JavaESB v1
JavaESB v1 KeyGen
 
Other Info
The Author
AI Books
Expert System Books
 
Other Products
Digi-Sign
 

Devise Knowledge Engineer Questions
The questions devised so far have been aimed at the user, i.e. the person using the system and wanting to be advised. The ESB questionaire is also presented to the domain expert or knowledge engineer (you normally) when using the ESB Knowledge Acquisition Program. With this program the domain expert populates the system with fuzzy data.

The question presented to the knowledge engineer needs to be asked in a different way. Consider when the domain expert enters data for a dog into our pet shop system (see SECTION 2 later). He needs to consider how each of the available options relates to (adds to or detracts from a dog being the recommended solution). For example our 1st question would be better phrased as follows (the options remain the same)

Q1. How would this animal be suited to people with the following fitness levels
a) Very fit, I get exercise every day and live for keeping fit
b) Fit, I exercise often and enjoy keeping fit
c) Average, I exercise when I get around to it but can do most activities normally
d) Unfit, I don't keep fit at all and get tired easily
e) Very unfit, I'm the architipal couch potato

We all know that dogs NEED exercise and so a fit person is more suited to having one and someone that sits at home all day would be less suited. Likewise Question 2 would be better phrased:

Q2. Asked about improving fitness users answer as follows. How would this pet suit each?
[A] Yes.
[B] No.
[C] I don't know.

Knowledge Engineer questions are entered on this part of the ESB Question Editor screen.

Question Types
In this section the term 'record' refers to a possible solution entered by the KE using the ESB KA program. The term 'record' and 'possible solution' should be considered interchangable.

Question Types are tricky and are down to your judgement as the system designer. The 'Question Type' is set by selecting the 'Multiple' or 'Single' radio button next to the Knowledge and System User Questions. These buttons govern the manner in which the Knowledge Engineer and System User respond to questions in the ESB Knowledge Acquisition (KA) and ESB User Interface (UI) Programs. They are either presented with slider bars or check boxes (as below):

Multiple   Single
or

So what does that mean and what effect does it have? If you're ahead of me you'll realise there are 4 possible combinations of Knowledge Engineer Question (KEQ) and System User Question (UserQ). The combinations are explained below:

1. KEQ = Single and UserQ = Single
You have decided that only one of the possible answers (for this question) should add to the likelihood of any record being the probable answer and you want the user to give only one positive answer. The KE sets the appropriate answer (for each possible solution) in the ESB KA program. The user makes a selection on the ESB UI program. All possible solutions in the system that match the users selection have their probability of being the actual solution increased, those that don't have it decreased.

Q6. Can you (normally) pick up and cuddle this pet? (KEQ)
Q6. Do you want to be able to pick up and cuddle your pet?
(UserQ)
[A] Yes.
[B] No.

2. KEQ = Multiple and UserQ = Single
You have decided that one or more of the possible answers (for this question) should add to the likelihood of any record being the probable answer but you still want the user to give only one positive answer. The KE sets (using the slider bars) whether each of the possible answers is positive or negative (for each possible solution) in the ESB KA program. The user makes a single selection on the ESB UI program. The setting made by the KE for the users actual response is then used to increase or decrease the probability of each record in the system being the actual solution. Consider the following:

Q1. How would this animal be suited to people with these fitness levels? (KEQ)
Q1. How fit and active are you?
(UserQ)
[A] Very fit, I exercise every day and live for keeping fit.
[B] Fit, I exercise often and enjoy keeping fit
[C] Average, I exercise when I get around to it but can do most activities normally.
[D] Unfit, I don't keep fit at all and get tired easily.
[E] Very unfit, I'm the archetypal couch potato
.

  For a DOG the KE would make settings as left. This means that if a user answered [A] then the probability of DOG being the solution is increased by the maximum. If the user answered [E] then probability of DOG being the solution is decreased by the maximum. Answers [B] and [D] change the probability by varying degrees. [C] is neutral.

3. KEQ = Single and UserQ = Multiple
You have decided that only one of the possible answers (for this question) should add to the likelihood of any record being the probable answer but you want the user to rate (or give a preference) for each of the possible answers.

None of the questions in our example fall into this category and so I will use a different example. Consider someone looking to buy a used car and an ESB system to find a car for them, a question might be:

Q1. What colour most closely matches is this particular car? (KEQ)
Q1. What colours of car would you consider buying?
(UserQ)
[A] Red.
[B] White.
[C] Blue.
[D] Black.
[E] Green.
[F] Silver.
[G] Yellow.
[H] Other

The car is (normally) only one colour and so the KE will select one of the options when entering data. The user is presented with a set of sliders and rates how he 'likes' each colour. ESB looks at the colour set by the KE, for each record in the system, and then increases/decreases the probability of that record being the solution by the response the user gave for that colour.

4. KEQ = Multiple and UserQ = Multiple
Use this option when you want the KE to enter values for all options and also the user. The system calculates an increase/decrease in probability for each record (possible solution) based on how closely the users responses match those entered by the KE. Consider Q18 from our example:

Q18. What age groups would this animal be suited to? (UserQ)
Q18. Which of the following age groups best describe your childrens age?
(UserQ)
[A] 0 to 5 years.
[B] 5 to 10 years.
[C] 10 to 15 years.
[D] 15 to 20 years.
[E] 20+.

Question Help
This is straightforward. If you decide that you would like to expand on the exact meaning of a particular question, or give some background information (or explanation) to help the user decide then that is done here. Create an HTML document and reference it from ESB by pressing the 'Add Help' button. You get this dialog to enter the URL.

Dependencies
Sometimes you will only want to ask a certain question IF the users answers to previous questions dictate so. Consider our first two questions:

Q1. How fit and active are you?
[A] Very fit, I exercise every day and live for keeping fit.
[B] Fit, I exercise often and enjoy keeping fit
[C] Average, I exercise when I get around to it but can do most activities normally.
[D] Unfit, I don't keep fit at all and get tired easily.
[E] Very unfit, I'm the archetypal couch potato
.

Q2. You don't consider yourself to be very fit but would you like to be?
[A] Yes.
[B] No.
[C] I don't know.

We only want to ask Q2 if a user responds with [D] or [E]. To do this select the 'Create Reliance' button on the Question Editor program. We want either of these options to lead to Q2 being asked. This is the screen, hopefully it's self explanatory.

It's possible to make a question dependent upon the answers to more than 1 previous question. In our example Q15 is also dependent upon Q14.

Importance
When considering the solution to a problem an expert does not assign the same importance to all of the facts he has gleaned about a particular problem. For example, there would be little point in recommending that someone buy a Rolls Royce (or Ferrari) when that person has said that their budget is £20K ($30K). In such as system the budget someone has should have a large bearing on the proposed solution. To assign an importance to questions in our system select one of the radio buttons at the bottom of the screen (default is normal).

For our 'Pet Shop Pet Selector' problem the questions that have most bearing on the final outcome is a matter of judgement (i.e. yours as the KE). I would suggest that Q1 (how active someone is) and Q8 (how much space do they have) are key questions that should have a greater impact. The decision is yours.

Finally
When all questions have been created, question types, importance, help and question dependencies assigned then save the system using the 'Save Questions' button.

Page:Previous 1 2 3 4 Next

   

Copyright(C) Paul C Caswell 1996-2008 All Rights Reserved

Flat 2, 25 Portman Crescent,
Bournemouth,
Dorset,
UK.