Adaptive Learning Systems: A review of Adaptation.
-
Upload
peter-brusilovsky -
Category
Technology
-
view
377 -
download
6
description
Transcript of Adaptive Learning Systems: A review of Adaptation.
Adaptive Learning Systems ���Towards “Adaptation Engine”
Peter Brusilovsky School of Information Sciences University of Pittsburgh, USA
Caveat Emptor
Overview
• Adaptation Technologies (what can be adapted and how) – Origins – Review – Place in the “Big Picture”
• How it could be implemented – “adaptation engine”
Key Aspects of Adaptive Systems
• Adapting to what? – User knowledge – User interests – User individual traits
• What can be adapted? – Adaptive sequencing of educational tasks – Adaptive content presentation – Adaptive ordering of search results
Technologies: The Origins
• Pre-Web AES Technologies
– ITS Technologies
– AH Technologies
• Web Technologies
• Post-Web Technologies • Brusilovsky, P. and Peylo, C. (2003) Adaptive and intelligent Web-based educational systems.
International Journal of Artificial Intelligence in Education 13 (2-4), 159-172.
Pre-Web Technologies
Adaptive Hypermedia Systems Intelligent Tutoring Systems
Adaptive Hypermedia
Intelligent Tutoring
Adaptive Presentation
Adaptive Navigation Support
Curriculum Sequencing
Intelligent Solution Analysis
Problem Solving Support
Pre-Web Technologies
• Intelligent Tutoring Systems – course sequencing – intelligent analysis of problem solutions – interactive problem solving support – example-based problem solving
• Adaptive Hypermedia Systems – adaptive presentation – adaptive navigation support
How to Model User Knowledge
• Domain model – The whole body of domain knowledge is
decomposed into set of smaller knowledge componens (skills, concepts, topics, etc)
• Student model – Overlay model
• Student knowledge is measured independently for each knowledge unit
– Misconceptions (bugs)
Simple overlay model
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N yes no
no
no yes
yes
Simple overlay model
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N yes no
no
no yes
yes
Weighted overlay model
Concept 1
Concept 2
Concept 3
Concept 4
Concept 5
Concept N 10 3
0
2 7
4
Bug models
Concept A
Concept B
Concept C
• Each concept/skill has a set of associated bugs/misconceptions and sub-optimal skills
• There are help/hint/remediation messages for bugs
Course Sequencing
• Oldest ITS technology – SCHOLAR, BIP, GCAI...
• Goal: individualized “best” sequence of educational activities – information to read – examples to explore – problems to solve ...
• Curriculum sequencing, instructional planning, ...
ELM-ART: Exercise Sequencing
Web
er, G
. and
Bru
silo
vsky
, P. (
2001
) ELM
-AR
T: A
n ad
aptiv
e ve
rsat
ile s
yste
m fo
r Web
-bas
ed in
stru
ctio
n. In
tern
atio
nal
Jour
nal o
f Arti
ficia
l Int
ellig
ence
in E
duca
tion
12
(4),
351-
384.
Beyond Sequencing: Generation
Kum
ar, A
. (20
05) G
ener
atio
n of
pro
blem
s, a
nsw
ers,
gra
de
and
feed
back
- ca
se s
tudy
of a
fully
aut
omat
ed tu
tor.
AC
M
Jour
nal o
n E
duca
tiona
l Res
ourc
es in
Com
putin
g 5
(3),
Arti
cle
No.
3.
Adaptive Problem Solving Support
• The core of Intelligent Tutoring Systems • From diagnosis to problem solving support • Low-interactive support
– intelligent analysis of problem solutions • Highly-interactive support
– interactive problem solving support
Intelligent analysis of problem solutions
• Intelligent analysis of problem solutions • Support: Identifying misconceptions (bug
model) and broken constraints (CM) • Provides feedback adapted to the user model:
remediation, positive help • Low interactivity: Works after the (partial)
solution is completed • Examples: PROUST, ELM-ART, SQL-Tutor
Example: ELM-ART
Interactive Problem Solving Support
• Classic System: Lisp-Tutor • The “ultimate goal” of many ITS developers • Several kinds of adaptive feedback on every step
of problem solving – Coach-style intervention – Highlight wrong step – What is wrong – What is the correct step – Several levels of help by request
Example: WADEIn
http://adapt2.sis.pitt.edu/cbum/ Bru
silo
vsky
, P. a
nd L
obod
a, T
. D. (
2006
) WA
DE
In II
: A c
ase
for a
dapt
ive
expl
anat
ory
visu
aliz
atio
n. In
: M. G
oldw
eber
and
P. S
alom
oni (
eds.
) Pro
ceed
ings
of
11t
h A
nnua
l Con
fere
nce
on In
nova
tion
and
Tech
nolo
gy in
Com
pute
r Sci
ence
E
duca
tion,
ITiC
SE
'200
6, B
olog
na, I
taly,
Jun
e 26
-28,
200
6, A
CM
Pre
ss, p
p. 4
8-52
.
Example-Based Technologies • While focused on problem solving, ITS research
developed several adaptive example-based learning approaches
• Example-based problem solving support – Adaptively suggesting relevant examples for given
problem and student state of knowledge (ELM-ART) • Adaptive worked out examples
– Steps could be presented with different level of details (fading with knowledge growth)
– Example steps could be replaced with problem steps
Adaptive hypermedia
• Hypermedia systems = Pages + Links
• Adaptive presentation
– content adaptation
• Adaptive navigation support
– link adaptation
• Could be considered as “soft” sequencing
– Helping the user to get to the right content
Adaptive navigation support • What could be done with links to provide
personalized guidance? • Direct guidance • Restricting access
– Removing, disabling, hiding • Link Ranking • Link Annotation • Link Generation
– Similarity-based, interest-based
Adaptive Annotation: InterBook
1. Concept role 2. Current concept state
3. Current section state 4. Linked sections state
4
3
2
1
√"
InterBook system
Adaptive Annotation: NavEx
Yudelson, M. and Brusilovsky, P. (2005) NavEx: Providing Navigation Support for Adaptive Browsing of Annotated Code Examples. In: C.-K. Looi, G. McCalla, B. Bredeweg and J. Breuker (eds.) Proceedings of 12th International Conference on Artificial Intelligence in Education, AI-Ed'2005, Amsterdam, the Netherlands, July 18-22, 2005, IOS Press, pp. 710-717
Adaptive Text Presentation���in PUSH (stretchtext)
Höö
k, K
., K
arlg
ren,
J.,
Wæ
rn, A
., D
ahlb
äck,
N.,
Jans
son,
C. G
., K
arlg
ren,
K.,
and
Lem
aire
, B. (
1996
) A g
lass
box
app
roac
h to
ada
ptiv
e hy
perm
edia
. Use
r Mod
elin
g an
d U
ser-A
dapt
ed In
tera
ctio
n 6
(2-3
), 15
7-18
4.
Adaptive Animation in WADEIn
Adapting to Individual Traits
• Source of knowledge – educational psychology research on individual
differences • Known as cognitive or learning styles
– Field dependence, wholist/serialist (Pask) – Kolb, VARK, Felder-Silverman classifiers
Style-Adaptive Hypermedia
• Different content for different style – Pictures for visually oriented – Little success, a lot of negative evidence
• Better idea: different interface organization and navigation tools for different styles – Adding/removing maps, advanced organizers,
etc.
Example: AES-CS
Interface for field-independent learners Tria
ntaf
illou
, E.,
Pom
port
is, A
., an
d D
emet
riadi
s, S
. (20
03) T
he d
esig
n an
d th
e fo
rmat
ive
eval
uatio
n of
an
adap
tive
educ
atio
nal s
yste
m b
ased
on
cogn
itive
sty
les.
Com
pute
rs a
nd E
duca
tion,
87-
103.
Example: AES-CS
Interface for field-dependent learners
Web Impact: Early Migration
• Intelligent Tutoring Systems (since 1970) – CALAT (CAIRNE, NTT) – PAT-ONLINE (PAT, Carnegie Mellon)
• Adaptive Hypermedia Systems (since 1990) – AHA (Adaptive Hypertext Course, Eindhoven) – KBS-HyperBook (KB Hypertext, Hannover)
• ITS and AHS – ELM-ART (ELM-PE, Trier, ISIS-Tutor, MSU)
Technology Fusion Adaptive Web Adaptive Educational
Systems
Adaptive E-Learning
Web Age Technologies
Adaptive Hypermedia Systems Intelligent Tutoring Systems
Information Retrieval
Adaptive Hypermedia
Adaptive Information
Filtering
Intelligent Monitoring
Intelligent Collaborative
Learning
Intelligent Tutoring
Machine Learning, Data Mining
CSCL
Native Web Technologies • Availability of logs
– Log-mining – Intelligent class monitoring – Class progress visualization
• One system, many users - group adaptation! – Adaptive collaboration support
• Web is a large information resource - helping to find relevant open corpus information – Adaptive content recommendation
Adaptive Collaboration Support
• Peer help / peer finding • Collaborative group formation • Group collaboration support
– Collaborative work support – Forum discussion support
• Awareness support
Educational Recommenders
• Motivated by research on IR and Recommender Systems
• Content based recommender systems • Collaborative recommender systems • Social recommender systems (based on
social links) • Hybrid Recommenders
Modeling User Interests
• Concept-level modeling – Same domain models as in knowledge
modeling, but the overlay models level of interests, not level of knowledge
• Keyword-level modeling – Uses a long list of keywords (terms) in place of
domain model – User interests are modeled as weigthed vector
or terms – Originated from adaptive filtering/search area
How it Fits Together?
Popular View on Adaptive Learning: Big PIcture
• A learning course (system) is an organized collection of learning content (objects)
• Students learn by moving from one content item to another interacting with each one depending on item nature (watch a movie, answer a quiz)
• Results are stored and used for learner modeling and analytics
A View on Adaptive Learning
• Adaptive learning could be achieved by adaptively selecting the next best content
• The job of adaptation engine is to use data about student (obtained before and during the course) to suggest next content item
What is (Partially) Correct • This is a valuable adaptation context, exactly the
place to use adaptive sequencing • Sequencing is an effective adaptation approach,
comes in several well-explored brands: – Mastery learning – Remedial sequencing – Proactive sequencing
• But – any personalized guidance technology that can guide the learner to the most appropriate content could be used in this context and there are other ways to do it – Adaptive navigation support – Recommendation with a ranked list
Lessons Learned I • Approaches that combine system-driven
adaptation with user ability to select content work better for “mature” learners that purely system-driven “Deus ex machina” approaches while sequencing is critical for younger kids – If you want to apply sequencing, consider other
guidance approaches as well • There are other approaches to support self-
regulated learning related to adaptation and they work really well – open learner model! – If you build learner model, make it open!
• Thanks, David, for explaining why we need it!
Exercise area
List of annotated links to all exercises available for a student in the current course grouped into topics
QuizGuide = OLM + ANS
• Topic-based Topic-based+Concept-Based
Concept-based vs Topic-based ANS
Lessons Learned II
• The largest impact is achieved by personalized guidance to complex activities (i.e., problems), while juggling static content has low impact – If you focus on sequencing, make sure you
have advanced learning content • Selection of activities based on learning
style is not (yet) an efficient approach, – If you want to build style-based adaptation, use
more complex approaches
What is Usually Missed • Learning objects are not necessary static files • Most efficient learning “content” is interactive (might
not even look like content, stored in files, copied) – Interactive simulations – Worked-out examples – Problems
• This is exactly the place to apply “within-content” adaptation – All kind of problem-solving support “tutors” – All kinds of adaptive presentation such as adaptive
animation and examples • There is a place for adaptation even beyond content
– Adaptive collaboration support
Lessons Learned III • Within-content adaptation is important
– Adaptive presentation significantly increases comprehension while decreasing learning time
– Provides vital problem-solving support where students needs most help
– Engages learners in interactive activities • There is no “single place” for adaptation
– Every type of content might use different approaches for adaptation and use own appropriate “engine”
– Different engines might need different information about learner and on different granularity levels
• ITS is a great technology for content-level adaptation, but existing monolithic ITS should be re-engineered to fit the traditional learning architectures
Requirements for AL architecture
• Support adaptation on several levels – Adaptive guidance (item to item) – Within-item adaptation – Adaptation beyond “items”, i.e., collaboration
• Data for learner modeling should be collected from all kinds of interactions
• Learner model produced from this data should be available for all components
ADAPT2 Architecture Portal
Activity Server
Student Modeling Server
Value-added Service
Brusilovsky, P. (2004) KnowledgeTree: A distributed architecture for adaptive e-learning. In: Proceedings of 13th International World Wide Web Conference, WWW 2004, New York, NY, 17-22 May, 2004, ACM Press,
User modeling server CUMULATE���
Event Storage
Inferenced UM
UM requests
Application ExternalInference Agent
InternalInference Agent
UM updates
Event reports
Event requests
All Pieces in Place?
Next Challenges: Architecture
• Post-Web Learning technologies are more diverse, but we need to find how to fit them into the architecture
• Educational games • Virtual and Augmenter Reality • Mobile learning • “Real World” learning
Next Challenges: Adaptation
• Most of existing adaptation technologies are based on knowledge engineering – Cognitive analysis – Metadata indexing
• Works well, but expensive • How we could use large volume of data
collected from many students to deliver and improve adaptation?
Social Personalization for AES
• Starting with technologies based on shallow processing of social data
• Social navigation support for open corpus resources – Knowledge Sea II
• Open Social Student Modeling with Social guidance – Progressor – MasteryGrids
Knowledge Sea II
Farzan, R. and Brusilovsky, P. (2005) Social navigation support through annotation-based group modeling. In: L. Ardissono, P. Brna and A. Mitrovic (eds.) Proceedings of 10th International User Modeling Conference, Berlin, July 24-29, 2005, Springer Verlag, pp. 463-472, also available at http://www2.sis.pitt.edu/~peterb/papers/FarzanBrusilovskyUM05.pdf.
Progressor
Hsiao, I. H., Bakalov, F., Brusilovsky, P., and König-Ries, B. (2013) Progressor: social navigation support through open social student modeling. New Review of Hypermedia and Multimedia 19 (2), 112-131.
MasteryGrids
Loboda, T., Guerra, J., Hosseini, R., and Brusilovsky, P. (2014) Mastery Grids: An Open Source Social Educational Progress Visualization. In: S. de Freitas, C. Rensing, P. J. Muñoz Merino and T. Ley (eds.) Proceedings of 9th European Conference on Technology Enhanced Learning (EC-TEL 2014), Graz, Austria, September 16-19, 2014.
The Challenge for Social Personalization
• Use large volume of learner community data to build more advanced adaptation approaches to replace or enhance “content-based” adaptation
• Example: Finding latent groups, meta-adaptation