Appendix F 22 May 11-50 Reasoning at AtGentSchool (Ontdeknet).ppt

Download Report

Transcript Appendix F 22 May 11-50 Reasoning at AtGentSchool (Ontdeknet).ppt

Reasoning in AtgentSchool
From conceptual framework to implementation
Events  reasoning  intervention
Koen Molenaar – David Kingma
Oxford, 22-5-2005
Contribution – WP1
 Deliverable 1.1:
 Descriptions of Ontdeknet
 Cases of usages within Ontdeknet
 English version Ontdeknet
 Deliverable 1.2:
 Intervention model in the learning proces
(Regulate, cognitive, meta-cognitive)
 Cognitive load theory
 Attention & Motivation
Contribution – WP2
 Deliverable 2.1:
 WOZ – insights in events
 Clarifying relation general and specific components
 Deliverable 2.2:
 Analysis of events – interventions within Ontdeknet
 Relation conceptual framework – Ontdeknet
 Deliverable 2.3
 Set up a general intervention model for Ontdeknet
 Validation intervention model with WOz
 Deliverable 2.4
 Experiences implementing conceptual framework events
Contribution – WP*
 WP3 Deliverable 3.1:
 Implementation conceptual framework events in Ontdeknet
 WP4 Deliverable 4.1:
 WOZ – formative evaluation setup.
 WP5
 Czech version Ontdeknet
 Teacher presentation Ontdeknet
 WP6
 Atgentive on national seminars
 WP7
 WP3 Kick off – contributions, tasks and planning
What did we do?
App lication
(Ontdeknet, ICDT)
….
User
Propos e
interven tion
user's activity
in application
environment
Retrieve
possible
alternative
focus
Select time
and mo de of
presentation
Atgentive agents
Preferences
Alternative foci
User information
User model
É
Why reasoning?
 We need to connect the event information to
interventions.
 Asking questions to the user is:
 distracting and disturbs the learning process and
 user might not provide reliable information.
 Adapt to the amount of available data, which
depends on the application / environment.
 Allowing the atgentive module to adapt the
application to the user.
Been there, done that
Lots of theory available on reasoning, using:
 Rules
 Agents
 Decision trees
 Neural networks
 ….
So what’s the problem?
 The Atgentive < > application interface is
unclear
 All applications are different and thus have
different kinds of data available
 Do we need planning for automated metacognitive support?
 Can we parameterize all the needed
information?
A few examples
 Which external event is important to the user,
and when should we notify him?
 How to decide on the modality of the
interaction?
 Limited resources available, how to decide if
and what other task to propose?
 When to re-attract an idle-user attention (when
is a user idle?)
Information available
A standalone Atgentive module cannot cope
with all the information available from the
application.
For example:
 Cognitive load of the user.
 Planning of the user and other resources
needed for certain tasks
 Actions of the application
Information available (II)
 Should the application pre-process the data?
 Should the application be able to somehow
adjust the reasoning parts in the Atgentive
module?
 Should the application have the last word?
WOz
• The Wizard of Oz construction allows to:




Validate Ontdeknet events (WP4)
Validate C.F. events (WP2)
Validate intervention model (WP4)
Validate the separation between application and the
Atgentive module
 Validate reasoning rules, before coding them
 Collect statistical data?
More details about the WOz in the afternoon!