The document proposes a hybrid recommendation approach for educational resource repositories that combines personalized and diversity-aware strategies. It includes a reactive strategy that combines long-term and short-term learning goals, a proactive strategy that fosters diversity, and a collaborative recommendation strategy. It also discusses the required knowledge including a domain ontology, learning objects, student profiles, and preference data. Finally, it outlines open issues regarding evaluating recommender systems in educational domains.
Personalized and diversity recommendation strategies for educational resources
1. Personalized and diversity-aware recommendation
strategies for educational resources
Almudena Ruiz Iniesta
Mercedes Gómez Albarrán
Guillermo Jiménez Díaz
Dpt. of Software Engineering and Artificial Intelligence
Complutense University of Madrid
2. Outline
Motivation
Required knowledge
Cascade hybrid approach
The reactive strategy: combining long-term and short-term learning goals
A proactive strategy that fosters diversity
A collaborative recommendation strategy
Open Issues
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
3. Outline
Motivation
Required knowledge
Cascade hybrid approach
The reactive strategy: combining long-term and short-term learning goals
A proactive strategy that fosters diversity
A collaborative recommendation strategy
Open Issues
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
4. Motivation
The development of electronic repositories for educational resources
has been intensified
High number of Learning Objects (LOs) exists in these repositories
Recommender systems support users in pre-selecting information
they may be interested in
We propose a recommendation approach for repositories of
LOs that adapts to the student learning profile
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
5. Motivation
Personalization
Cascade Hybrid Recommender Reactive Proactive
strategy strategy
Diversity
Case-based strategy
Collaborative
strategy
Collaborative strategy Learning community
opinion
Obtain recommendations even when
a LO has only few ratings
a student is new in the repository
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
6. Outline
Motivation
Required knowledge
Cascade hybrid approach
The reactive strategy: combining long-term and short-term learning goals
A proactive strategy that fosters diversity
A collaborative recommendation strategy
Open Issues
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
7. Required knowledge: domain ontology
The ontology is populated with concepts in the field of study
Ontologies provide a general indexing scheme that lets include
similarity knowledge between the concepts representing the
domain topics
The ontology should also establish a precedence property among
the concepts
The precedence helps to establish the learning paths
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
8. Required knowledge: Learning Objects
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
9. Required knowledge: student profile
The goals achieved in the learning process
Concepts that the student should know and the mastery level
achieved in each of them
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
10. Required knowledge: preference repository
The preference repository
Rating scores explicitly assigned by the students to each LO
The profile that the student had when she rated the LO
Evolving nature of the profile
The collaborative recommendation strategy looks for students with similar
mastery level
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
11. Outline
Motivation
Required knowledge
Cascade hybrid approach
The reactive strategy: combining long-term and short-term learning goals
A proactive strategy that fosters diversity
A collaborative recommendation strategy
Open Issues
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
12. Cascade First-stage: Case-based strategies
Two alternative, each of them Reactive Proactive
satisfies an essential strategy strategy
requirement of recommenders
of educational resources
Collaborative
Personalization strategy
Reactive strategy: combining long-
term and short-term learning goals
Overspecialization
Proactive strategy that fosters
diversity
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
13. The reactive strategy: combining long-term and short-term
learning goals
Student query
{C1,1; C2,1… }
Retrieve LOs that
cover same or
similar concepts
Retrieval step Filter LOs
not ready to be
explored
Rank according to
Ranking step the LO Quality
Ranked list of
recommended
LOs
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
14. The reactive strategy: combining long-term and short-term
learning goals
The retrieval step
Concepts “ready to be explored”
Concept already explored by the student competence level
It is a concept that the student has not explored yet but she can discover it
if a concept c1 precedes a concept c2 in the ontology, a student can discover c2
if the student competence level for c1 exceeds a given “progress threshold”
Concept unreachable
Concepts already explored Concept ready to be
discovered
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
15. The reactive strategy: combining long-term and short-term
learning goals
The ranking step
Quality: weighted sum up of two relevancies
Similarity between the query concepts and the concepts that the LO covers
Pedagogical utility
Strategy that promotes filling the student’ knowledge gaps
High values of PU indicates that the student has a poor knowledge of most of
the concepts covered by the LO
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
16. A proactive strategy that fosters diversity
Propose a set of LOs according
to the student profile
Do it
Student END proactive
selects a LO recommendation
More like this
Refine the recommendation
according to the student feedback
Yes Can refine the No
recommendation?
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
17. A proactive strategy that fosters diversity
Navigation-by-proposing
Diversity Select LOs from different partitions
in the space of LOs
First stage
Reinforce vs. Discovering
Ranking
Priority to LOs that maximize the number of
covered concepts
Reinforce Discover
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
18. A proactive strategy that fosters diversity
Second stage
Breadth-first traversal of the taxonomy
One partition for each concept in the current level
Ranking
Priority to LOs that maximize the number of covered concepts
In case of reinforcement, clusters in which the student has shown a low
mastery level
Repeat until leaves are reached
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
19. Cascade Second-stage: A collaborative recommendation
strategy
Select the LOs provided by the case-based
recommender with higher interest according to
the opinions of other members of the community
User-based nearest neighbour
Neighbourhood formation
candidates: students who rated the LOs proposed
by the case-based recommender
similarity between the target student profile and
the profile that the neighbour had when she rated
the LO
Rating prediction and top-k selection
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
20. Outline
Motivation
Required knowledge
Cascade hybrid approach
The reactive strategy: combining long-term and short-term learning goals
A proactive strategy that fosters diversity
A collaborative recommendation strategy
Open Issues
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
21. Open Issues
Recommender systems in the learning domain impose new
challenges in the evaluation process
More important to measure the impact of the recommender in the final
user
Our evaluation
Goal-Questions Metrics method
Analyzing the usability of the repository
Analyzing students’ grades
Analyzing the impact of different recommendation strategies
Repository of LOs for Computer Programming
Basili VR, Rombach HD. The TAME project: towards improvement-oriented software environments. Software Engineering,
IEEE Transactions on. 1988;14(6):758-773
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources
22. Thank you!
Personalized and diversity-aware recommendation
strategies for educational resources
Almudena Ruiz-Iniesta
almudenari@fdi.ucm.es
http://gaia.fdi.ucm.es/people/almudena
RecSys 2010 - Doctoral Symposium: Personalized and diversity-aware recommendation strategies for educational resources