Présentation du projet REFRER sur les référentiels de ressources éducatives r...
Rec systel 2012 competency based recommendation
1. RecSysTEL Workshop 2012
Saarbruecken September 18, 2012
Competency Comparison Relations for
Recommendation in Technology
Enhanced Learning Scenarios
Gilbert Paquette, Délia Rogozan, Olga
Marino
www.licef.ca/cice
Canada Research Chair in Instructional and Cognitive
Enginerring (CICE)
LICEF Research Center
2. Background
Add semantic references to scenario
components: actors, tasks and resources to
educational modeling languages such as IMS-LD
(2003)
– Paquette and Marino, 2005
“Include the improved modeling of users and
items, and incorporation of the contextual
information into the recommendation process”
– Adomavicus and Tuzhilin (2005)
The “Adaptive Semantic Web” opens new
approaches for recommenders systems: use of
folksonomies and ontological filtering of resources
– Jannach et al, 2011
3. Recommendation (assistance)
Principles
Epiphyte – grafted on the scenario process
but external to it; no scenario modification
Multi-agent system: agents are associated
to tasks at different levels in the scenario
Flexible association: one, some or all of the
tasks are assisted.
Delegation between a task agent towards its
super tasks agents; tree topology
5. The implemented recommender
model
Recommender = {rules}
Rule = <actor, event, condition, action >
Event =
– Activity transition (started, terminated, revisited,…)
– Time spent (activity, global …)
– Resources opened, reopened,…
Condition = boolean expression comparing:
– Target actor progress in the scenario + knowledge and
competencies acquired + evidence => User persistent model
– Resources: prerequisite and target competencies
– Activities: prerequisite and target competencies
Action = advice, notification, model update
6. Semantic Referencing of
Resources
Of what
– Actors, activities, documents, tools, models, scenarios …
Why
– Help select resources at design time for better quality scenarios
– Inform users of the resources’ content at design or delivery time
– Assist users according to their knowledge and competencies
How
– Associate formal semantic descriptors to resources from a
domain ontologies and/or competencies based on ontology
references
7. Knowledge Descriptors
Classes and instances
(From OWL-DL domain ontologies)
General properties:
–Domain – Data Properties
–Domain – ObjectProperty – Range
Instanciated properties (facts):
–Instance – Property
–Instance – Property – Value
8. Competency Descriptors
Knowledge descriptors
Competency descriptors
– (K, S, P) triples
K: Knowledge descriptor
K=Planet
K=Planet
– From a OWL domain ontology
S: Generic Skill
S=Apply
S=Apply
– From a 10-level taxonomy (Paquette, 2007)
P: Performance level
P=Expert
P=Expert
– A combination of P-values (Paquette, 2007)
9. Referencing Process in the
TELOS Implementation
Ontology Resource Semantic
1
1 contruction 2
2 selection 3
3 Referencing
or import Of resources
… and/or
competencies
10. Semantic Search Methods
Type de recherche Type de résultat
Simple Exact match
Using key words from the ontology
Advanced Exact match
Using knowledge and competency Semantically
boolean query near match
Exact match
Resource Pairing
Semantically
Using semantic comparison between
queried ressource and other resources near match
→ Rests on knowledge and competency comparison
11. Knowledge Comparison
(K1 et K2)
Based on the structure of the ontology where the
knowledge descriptors are stored
Compare the neighbourhoods of K1 and K2
Possible results
– K2 near and more specialized / general than K1
12. Competency Comparison
C1=(K1, S1, P1) et C2=(K2, S2, P2)
Based on knowledge comparison (K)
Base on the distance between skills’ levels (H) and
performance levels distances(P)
Possible results
C2 veryNear / Near C1
C2 stronger / weaker than C1
C2 more specialized / general than C1
14. Competency comparison
within rule conditions
A competency-based condition is a triple:
– ObjectCompetencyList is the list of prerequisite or
target competencies of another actor, a task or a
resource to be compared with user’s actual
competency list
– Relation is one of the comparison relations :
Identical, Near, VeryNear, MoreGeneric, MoreSpecific,
Stronger, Weaker, or any combination of these.
– Quantification takes two values: HasOne or HasAll
EX: HasAll /NearMoreSpecific / Target competencies for Essay
EX: HasOne/Weaker/Target competency for Build Table activity
18. Achievements in this project
Extension of the TELOS Technical Ontology for
semantic referencing of resources, search and
recommendation
Definition of a Typology of semantic descriptors
(ontology descriptors and competenciers)
Search methods for resources ‘identical’ ou ‘near’
sémantically
Recommendation Model: based on competency
comparison between actors, tasks and resources
New integrated suite of tools: Semantic referencer,
Semantic search tools, Competency and Ontology
editors, Integration to recommanders scenarios,
Recomenders’ rule editor.
19. Future steps
More experimental validation to refine the semantic relations
between OWL-DL references, i.e adding weights to the various
comparison cases
Investigate recommendation methods for groups in
collaborative scenarios (permitted by our model of multi-actor
learning scenarios)
Improve the practical use of the approach, partly automate
tasks, improve the ergonomics
Investigate the integration of other recommendation methods
(e.g. user analytics)
“Free” the suite of tools from TELOS to extend its usability on
the Web of data.
20. RecSysTEL Workshop 2012
Saarbruecken September 18, 2012
Questions ?
Comments ?
Gilbert Paquette, Délia Rogozan, Olga
Marino
www.licef.ca/cice; www.licef.ca/gp
Canada Research Chair in Instructional and Cognitive
Enginerring (CICE)
LICEF Research Center
Notes de l'éditeur
Donner des exemples (à l ’oral) pour chaque type de recherche.
Voisinage ‘proche’ au sens qu’on ne descend pas la hiérarchie des classes, propriétés, etc.…