This document discusses using simulations to research emerging effects in learning networks. It outlines why simulations are useful given limitations of real learning networks. It also describes appropriate simulation frameworks, a methodology for designing simulations, and potential research focuses including exploring recommendation algorithms and measuring performance using learning theory and social network measures. Open questions are raised about how to observe and measure emergent effects, necessary statistical analysis, and integrating additional factors into simulations.

1. Discovering emerging effects in Learning Networks
with simulations
Hendrik Drachsler 12/17/2007

2. Agenda
1. Why using simulations for research in Learning
Networks?
2. Appropriated simulation frameworks
3. Methodology approach for designing simulations
4. Research focus on simulations for LNs
5. Expectations O
5E t ti Open questions
ti

3. Emerging Effects
Example of an emerged effect

4. 1. Why using simulations for research in
Learning Networks?
- Emerging behavior of learners
in LNs (Navigation support)
Why simulations:
- Limited availability of LNs
- Experiments are cost intensive
and limited in time amount
time,
of learners and UoLs
Especially research in emerging effects requires long term
perspectives and huge amount of learners. [ISIS Example]

5. 2.
2 Appropriated simulation frameworks
1. RePast
- Highlevel platform
(2. MASON application)
(programmable
bl li ti )
3. scape
4. Netlogo
- Swarm
5 StarLogo
-
5. TeamBots
6. Player/Stage
- Framework & library platforms
7. Breve
(conceptual frameworks)
8. StarLogo
9. MASON
- NetLogo
10. Processing
-R Repastt
11. MadKit
- Java Swarm
12. Cormas
13.
13 Magsy
14. Simpack

6. 3.
3 Methodology Approach for Simulations
Model Simulated d t
Si l t d data
Simulation
Abstraction Similarity
Data gathering
Target Collected data

7. 4.
4 Research focus on a LN simulation
Exploration of different kinds of bottom-up
p p
recommendation algorithm on different sized LNs.
1. User based
1 User-based filtering
2. Item-based filtering
3. Tag-based filtering

8. 4.
4 Research focus on a LN simulation
Measuring performance of three algorithms in three
different sized LNs on :
Classic Learning Theory Measures:
• Goal attainment
• Time to reach goal
• Dropout rate
Social Network Aspects:
• Connectivity (Exploration of the LNs through Learners)
y( p g )
• Centrality (importance of a Learner, count of the number of ties)
• Closeness (sum of the shortest distances learners)
• Variety of paths

9. Expectations Open Questions
Expectations:
p
- Conditions of LNs in which specific algorithms perform
better than others.
- An Evaluation approach for the combination of SNA
techniques with Learning Theory Measures
Open Questions:
- How can we observe / and measure what emerges?
- What kind of statistical analysis is needed?
- How to combine SNA measures with classic learning
research?
- How to integrated user tagging into a simulation?

10. References:
Journals for Simulation Research:
- Journal of Artificial Societies and Social Simulation (JASSS)
- Journal of Complexity International
- Journal Artificial Life
Mailing li t N
M ili lists / Newsletter:
l tt
- http://www.comdig.org/
- http://ec-digest.research.ucf.edu
- http://www jiscmail ac uk/lists/evolutionary-computing html
http://www.jiscmail.ac.uk/lists/evolutionary computing.html
- http://www.genetic-programming.org/gpmailinglist.html
Websites:
- http://www.multiagent.com
- http://cress.soc.surrey.ac.uk/s4ss/index.html (Simulations for Social
Scientists)
- http://www swarm org/wiki/Main Page
http://www.swarm.org/wiki/Main_Page