Intelligent Tutoring Systems (ITSs) constitute an alternative to expert human tutors, providing direct customized instruction and feedback to students. ITSs could positively impact education if adopted on a large scale, but doing that requires tools to enable their mass production. This circumstance is the key motivation for this work. We present a component-based approach for a system architecture for ITSs equipped with meta-tutoring and affective capabilities. We elicited the requirements that those systems might address and created a system architecture that models their structure and behavior to drive development efforts. Our experience applying the architecture in the incremental implementation of a four-year project is discussed.
Presentation on how to chat with PDF using ChatGPT code interpreter
201407 A System Architecture for Affective Meta Intelligent Tutoring Systems
1. A System Architecture for
Affective Meta Intelligent Tutoring Systems
Javier Gonzalez-Sanchez, Maria Elena Chavez-Echeagaray,
Kurt VanLehn, Winslow Burleson, Sylvie Girard, Yoalli Hidalgo-Pontet, and Lishan Zhang
2. A System Architecture for
Affective Meta Intelligent Tutoring Systems
Support ITS development and inclusion of
Affective and Meta-Tutoring Capabilities
Affective Meta-Tutor (AMT) Project(VanLehn et. al., 2014)
Step-based ITS + Meta-Tutor + Affective Companion
4-year research project
4. Support ITS Development | Related Work
§ ITS design patterns
(Harrer, Pinkwart, McLaren, & Scheuer, 2008; Devedzic & Harrer, 2005)
§ Reusable components and learning objects
(Koedinger, Suthers, & Forbus, 1999; Ritter, Blessing, & Wheeler, 2003;
Ritter & Koedinger, 1997)
§ Use of off-the-shelf tools as an integrated part of the ITS
development process
(e.g., Aleven, Sewall, McLaren, & Koedinger, 2006)
§ Authoring tools
(e.g., Murray, Blessing & Ainsworth, 2003)
1
6. AMT Project | Summary1
2009 2010 2011 2012 2013
ITS ITS ITS ITS
MetaTutor MetaTutor MetaTutor
ALC
§ Affective Meta Tutor (VanLehn et. al., 2014)
§ Schedule constraints
§ Robustness
§ One click installation
§ Work with and without network access
§ Multiplatform Tool
Summer Camps
High School Students
Several Components
7. AMT Project | Summary1
2009 2010 2011 2012 2013
ITS ITS ITS ITS
MetaTutor MetaTutor MetaTutor
ALC
§ Modifiability
§ Extensibility
§ Robustness
§ Cost-Effectiveness
• Undergrad students (developers)
• High turnover in the development team
Students graduate after
1 year (average)
Change is Expected
9. Approach
• Architectural Principles:
o encapsulation
o low coupling
o centralized shared data
o layering
• Component-Based and Pattern Oriented design:
o “divide and conquer”
o complex components are split into several collaborative
components
• Developers know their part
• Developers have constrained freedom
2
11. AMT 2009, version 0.1
• Interactive graphical
interface (student freely
manipulate nodes and
connect them)
• Task (problems to be
solved) are edited by
human experts (outside
the tutor)
2
12. AMT 2009, version 0.7
• Tutor, version 2009 release 0.7
• User can assign expressions to nodes and expressions are
parsed in real time to provide visual feedback
2
13. AMT 2009, version 0.81
• New look and feel of nodes and interconnection rules
• User can “run” and can create plots of the model
2
14. AMT 2009, version 0.91
• ITS runs the model
• Characteristics that
the model should
satisfy are verified as
questions in a Quiz
• Metatutor and ALC
becomes external
agents with internal
advocates
2
15. AMT 2010, version 1.11
• The system can run on 2 profiles: student and professor. The system supports
the authoring process.
• Competition flavor, where points are given for each correct step. Asking for
hint costs points.
• Companion with non- animated human face.
2
16. AMT 2010, version 1.9
• Increase student freedom: nodes has menus and color indicators.
• Fine-grain student support.
2
21. AMT
• 1,643 revisions posted to our SVN repository
• 8 released versions
• 16 packages
• 120 classes
• 1507 methods
• 1810 attributes
• 37,374 lines of code
• 2 third-party subsystems: a production-rule system and a third-party
implementation of emotion recognition algorithms were used to support
the application development
• 4 developers was working concurrently in every stage (15 developers in
total)
2
22. Evaluation
The structural software quality was evaluated using software metrics for size,
complexity, and coupling:
• number of packages (P),
• number of classes (F),
• number of functions (Fn),
• number of lines of code (LoC),
• number of comments (LoCm),
• average cyclomatic complexity (AvC),
• maximum afferent coupling (MaxAC), and
• maximum efferent coupling (MaxEC).
3
23. Evaluation
We report the evaluation of four AMT application releases, one
from each development year, as follows:
(1) Release 742 implemented the first deployed Tutor; coding the
skeleton of the system was the primary goal during this year.
(2) Release 1277 refashioned the User Interface and
implemented an enhanced Tutor.
(3) Release 1545 included a Meta Tutor, continued refashioning
the User Interface, and enhanced the Tutor module.
(4) Release 1643 added the Affective Companion capabilities,
enhanced the Meta Tutor, and refactored the User Interface
and Tutor.
3
25. Discussion4
Size measurements (P, F, Fn, LoC, and LoCm) show a correspondence of the
requirements implemented in each release and the size of the application, as well
as a balance in its granularity
26. Discussion4
The average complexity (AvgC) at the module level remains within acceptable
ranges – below 5; however, at a fine-grain level (classes), not shown in the table,
those values are not always acceptable.
27. Discussion4
The decrease in average complexity in the latest versions of Tutor shows the
refactoring outcome (functionality was fixed and developers focused on code
improvements).
28. Discussion4
Lower values in coupling measures (MaxAC and MaxEC) are better since they are
a sign of independence; the high values of coupling in Tutor can be justified
because they belong to the User Interface (highly connected); Meta Tutor values
are acceptable, but Affective Companion values suggest that a refactoring would
be required in the implementation of this module.
29. Summary
• Shared our experience developing an ITS with meta and
affective capabilities.
• Presented AMT system architecture.
• Defined requirements and qualities and have shown how the
AMT system architecture addresses them to support large-
scale reuse.
• Use software metrics for different releases of one AMT
application to show how the system architecture provided a
flexible partition of the system that facilitates modifiability,
extensibility, and integrability.
5
30. A System Architecture for
Affective Meta Intelligent Tutoring Systems
javiergs@asu.edu | helenchavez@asu.edu