This study examined whether student stereotypes could be used to improve predictions of problem-solving performance in MOOCs. The researchers tested simple stereotypes based on demographics and performance, but found they did not yield accurate models of different student groups' learning. More advanced stereotypes based on patterns of problem-solving behavior also did not distinguish groups with significantly different learning models. The findings suggest stereotypes may not effectively represent the finer-grained differences in how students approach learning. Overall, the study did not find evidence that stereotype models can improve predictions of problem-solving over alternative models of individual student learning.
Plant propagation: Sexual and Asexual propapagation.pptx
Stereotype Modeling for Problem-Solving Performance Predictions in MOOCs and Traditional Courses
1. Stereotype Modeling for Problem-
Solving Performance Predictions in
MOOCs and Traditional Courses
Roya Hosseini, Peter Brusilovsky,
Michael Yudelson, Arto Hellas
1
2. Background
2
Stereotype Models
Elaine Rich (1983)
Feature-Based Models
Corbett & Andreson (1995)
InterestGoal
Knowledge
High AccuracySimple
Leverage large
volumes of data
3. MOOC: A Comeback of Stereotypes?
3
Learning Analytics:
Define demographic
stereotypes, use data
to find differences
Success in MOOC drop-
out/failure prediction
Educational DM:
Mine stereotypes
from data
4. A Simple way to Adaptive MOOC?
4
Treat learners
differently based
on demography
Recognize stereotype
on the early stages and
treat differently
5. Research Focus
• Past work: Coarse-grain prediction and
adaptation (dropped-out – completed)
• Our goal: Finer-grain adaptation – content
sequencing, navigation support
• Use stereotypes to guide
different groups of
students along different
pathways
5
6. How We Can Do It?
• A data-driven approach to personal guidance:
– Use data to learn a student model that could be used
to predict student performance on a problem
– Choose a problem (or a topic) based on predicted
performance
• How to do it better with stereotypes?
– Use large volume of course problem-solving data
– Determine learning-related stereotypes
– Learn different models for different stereotypes
– Use stereotype-specific models for advanced data-
driven personalization
6
7. Research Context: A Programming MOOC
7
Test My Code Plugin
NetBeans IDE Compiled Correctness
✗
✓
✓
✓
0%
20%
80%
100%
Snapshots
• Problem-solving (coding)
o real context
o rich data
o automated feature analysis of solutions
11. Results for Simple Grouping
11
0
0.2
0.4
0.6
0.8
Gender Edu. #Trans. P.Solv. %Corr.
Suboptimal
Expected
Out of the five simple
grouping approaches,
only two had a non-zero
score.
12. Advanced Data-Driven Grouping
12
a
Processing
Snapshots
Mining Problem-
Solving Sequences
Grouping By
Behavior Profiles
1 2 3
A—L: long steps
(individualized)
> < =
>
=
<
0
b c
d e f
g h
j k l
i
Concept
Correctness
SPAM
_aaA_
_jjJgK_
Frequent patterns
(ac, _aa, jj_, ad, …)
Normalized vectors
of frequent patterns
C1 C2 C3 C4 C5
Hierarchical Spectral
Guerra, J., Sahebi, S., Lin, Y.-R., and Brusilovsky, P. (2014) The Problem Solving Genome: Analyzing Sequential Patterns of Student Work with Parameterized
Exercises. 7th International Conference on Educational Data Mining (EDM 2014), pp. 153-160.
14. Results for Advanced Grouping
14
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
C1 C2 C3 C4 C5
None of the advanced
grouping approaches
(C1 — C5) were scored
as ideal, as we have not
found at least two groups
that were voted as
sufficiently different by
the group models of
student learning.
Expected
15. Main Take-Away Points
15
o No discernable differences in
cross-prediction accuracies in
simple and advanced student
groupings
o Stereotype models might not be
good as alternative models to
represent finer-grain student
learning
16. Main Take-Away Points
16
o Finding a useful learning-focused
stereotype, like good students or
slow students is NOT trivial.
o There might be students who
approach learning differently, but
the distinction between these
approaches are orthogonal to the
conventional dimensions that we
apply to quantify learning.
17. Contributions
• (Past) Resource usage
• (This Work) Micro patterns in problem-solving
17
Behavior Analysis in MOOC
Code Submission Analysis
• Problem-solving patterns (behavior)
• Impact of patterns on student learning/performance
18. Relevant Readings
• Concept-based analysis of MOOC program submissions
– Yudelson, M., Hosseini, R., Vihavainen, A., and Brusilovsky, P. (2014)
Investigating Automated Student Modeling in a Java MOOC. In:
Proceedings of the 7th International Conference on Educational Data
Mining (EDM 2014), London, UK, July 4-7, 2014, pp. 261-264.
– Hosseini, R., Vihavainen, A., and Brusilovsky, P. (2014) Exploring
Problem Solving Paths in a Java Programming Course. In: Proceedings of
Psychology of Programming Interest Group Annual Conference, PPIG
2014, Brighton, UK, June 25-27, 2014, pp. 65-76.
• The Problem-Solving Genome approach
– Guerra, J., Sahebi, S., Lin, Y.-R., and Brusilovsky, P. (2014) The
Problem Solving Genome: Analyzing Sequential Patterns of Student Work
with Parameterized Exercises. In:Proceedings of the 7th International
Conference on Educational Data Mining (EDM 2014), London, UK, July 4-
7, 2014, pp. 153-160.
18
19. 19
* This work used the Extreme Science and Engineering Discovery Environment (XSEDE),
which is supported by NSF award OCI-1053575. Specifically, it used the Bridges system,
which is supported by NSF award ACI-1445606, at the Pittsburgh Supercomputing Center.
Questions?
Stereotype Modeling for Problem-Solving Performance
Predictions in MOOCs and Traditional Courses
R. Hosseini, P. Brusilovsky, M. Yudelson, A. Hellas
Notes de l'éditeur
Animation: we have to emphasize that one big reason to use stereotype modeling is that these models are simple.
Why we were motivated to use stereotype modeling? (1) first was that they were simple and scalable…but there are more reasons such as their good results at predicting drop-out and failure in MOOC, and also they could make MOOC adaptive.
Why we were motivated to use stereotype modeling? (1) first was that they were simple and scalable…but there are more reasons such as their good results at predicting drop-out and failure in MOOC, and also they could make MOOC adaptive.
In this work, we have attempted to explore the prospects of stereotypes in MOOCs beyond dropouts--- for predicting student performance at the problem level. We used data from programming MOOC that included a large share of problem-solving activities and provided fine-grained data about user problem-solving behavior. Our goal was to find stereotypes that could be useful for predicting user’s success at solving problems.
Now explain the data and its distinct characteristics.
Performance prediction is done on separate batches of “train” and “test” batches during 20 different runs of train-test.
Every point is a mean performance on “test” batches, Serifs show Standard Errors of the mean during the 20 runs.