Applying learner modelling for user interface assistance in simulative training systems Alexander H

1
Applying learner modelling for user interface
assistance in simulative training
systemsAlexander Hörnlein, Frank PuppeDept.
for Artificial Intelligence and Applied Computer
Science,JMU Würzburghoernlein,
puppe_at_informatik.uni-wuerzburg.de

• Gliederung
• Motivation
• Task domain
• System states and actions
• Overlay model
• Intervention system
• Discussion and future work

2
Motivation

• Simulative training systems have complex
  interfaces
• Students want to learn the content NOT the UI
• No one reads manuals
• Hard/Impossible to teach UI ex-cathedra with
  distributed groups/individuals with asynchronous
  access
• Learners of different competence need individual
  help

3
Applied kinds of help

• Comprehensive system metaphor
• Static online courses with basic help
• (Dynamic context-sensitive) built-in help
• Additionally
• Active help-system
• Learner modelling
• Intervening wrong user actions
• Providing help for recent error(s)

4
Task domain

• Learner can freely switchbetween main session
  tasks
• Tasks consist of (sub-)tasksor atomic actions

diagnose
switch to diagnose mode
delete wrong diagnoses
navigate diagnoses tree
add diagnosis
rate diagnosis
click on Diagnose
click on delete
scroll to sub-tree
open sub-tree
click on diagnosis
click on established or suspected
click on
5
System states and actions

• System state is described with partial states
• Actions change the system stateTransition
  function
• System states influence available
  actionsAvailability function
• Task objectives result from system
  statesObjective function

6
Action lists

• Sequences of actions to reach the/a final state
  for a given task

click on established or suspected
click on delete
click on diagnosis
click on diagnosis
click on established or suspected
click on diagnosis
click on diagnosis
click on delete
click on delete
enter search text
click on hinzufügen
select diagnosis
click on established or suspected
click on Textsuche
click on Textsuche
select diagnosis
enter search text
click on hinzufügen
click on reset
7
Concepts

• Structural concepts
• Task decomposition to sub-tasks
• Knowledge about task domain
• High level order concepts
• When to start a task
• Knowledge about objective function
• Action concepts
• Action lists for a given task
• Knowledge about task domain (leaves), transition
  function and availability function

8
Overlay model

• Set of concepts
• Set of ordered symbolic values
• Interval
• Numerical score function
• Symbolic score function
• Inverted symbolic score function
• ? An overlay model is the 6-Tupel

9
Changes of the model
set of all functions
set of all change functions
10
Example

• Set of concepts C to diagnose switch to
  diagnose, change ,
• Set of symbolic values S N3, N2, N1, P0, P1,
  P2, P3
• Interval I -25,25
• m1 initially set to m1(c)0
• m2

• m3(n) N3, if N2, if N1, if P0, if P1,
  if P2, if P3, if

N3 N2 N1 P0 P1 P2 P3
-15 -10 -5 0 5 10 15
11
Rule sets

• Two rule sets to change overlay model
• State rulesIF expected diagnoses changedAND
  NOT action click on diagnoseTHEN DECREASE
  VALUE OF if expected diagnoses change then one
  should diagnose BY
• Dependency rulesIF VALUE OF to open diagnose
  subtree click on BELOWTHEN DECREASE VALUE
  OF to open therapie subtree click on BY

12
Intervention system requirements

• Intervention system must
• Prevent the learner from doing wrong actions
• otherwise the learner has to manually undo the
  last action,which is sometimes not possible
• Provide help if the learner seems to be stuck
• Be unobtrusive
• otherwise the learner cant focus on learning
  subject

13
Intervention system workflow

• on learner action system state and learner
  action (history) are gathered
• state rules and dependeny rules are executed
• if a concept has a bad rating (overlay model)an
  appropriate prepared intervention is returned
  (with a weight)(model?intervention rules)
• the intervention gets a score based on
• its weight
• the kind and number of interventions returned
  after recent learner actions
• the kind and number of interventions recently
  returned
• all interventions with a score below a certain
  threshold are held back
• if there are interventions left, then
• the intervention with the highest score is
  returned
• the rating (overlay model) of the interventions
  concept is increased(intervention?model rules)
• the learner action is cancelled
• the intervention content is displayed (by the
  feedback agent)

14
Intervention system
15
Discussion and future work

• Done
• Implementation nearly complete
• Rule sets for most concepts of all-but-one main
  task
• ToDo
• Complete implementation and rule sets
• Fine-tuning of rules and intervention score
  function
• Future
• Evaluation
• Preset with stereotypes
• Enable the learner to modify model
• Different agent
• Theoretical Define different types of task
  domain relations, conditions

16
Questions?