In the talk presented at last RecSys conference we discussed some of the common challenges in the e-learning industry that we are facing at Cloud Academy such as: heterogeneity of content to recommend and specific recommendation goals targeting the user training objectives.
5. Intent-based
User task - standard
Roberto Turrin
Personaliza0on Challenges in E-Learning
Watching a movie
Listening to a song
Planning a travel
I know what I want to achieve
I don’t know what to do
Explora0on level
Discovery
Goal-driven
Search
Watching a movie with my partner
Planning a travel with my family
I know how I want to get sth Watching the last movie of TaranPno
Finding the Pmetable of flights to Madrid
Standard
Enjoyment
6. Intent-based
User task - educaPon
Roberto Turrin
Personaliza0on Challenges in E-Learning
Studying something
I know what I want to achieve
I don’t know what to do
Explora0on level
Discovery
Goal-driven
Search
Learning Python
Preparing for a cerPficaPon
TesPng the level of knowledge
Mastering BBQ cooking
Becoming a data scienPst
I know how I want to get sth Doing an advanced course about
deep learning
Educa4on
Learning
7. User profile - interests
Standard
Roberto Turrin
Personaliza0on Challenges in E-Learning
Interests/tastes
Educa4on Interests/tastes
Comedy vs drama movies
Rock vs pop songs
Statues vs painPngs
Sea vs mountain vacaPon
Astrology
Machine Learning
What I am interested in
What I prefer
What I am interested in
What I prefer
Enjoyment
Learning
8. User profile - interests
Roberto Turrin
Personaliza0on Challenges in E-Learning
Educa4on Interests/tastes Astrology
Machine LearningWhat I am interested in
What I prefer
Learning
?
9. User profile - educaPon-specific
Standard
Roberto Turrin
Personaliza0on Challenges in E-Learning
Interests/tastes
Educa4on Interests/tastes
Comedy vs drama movies
Rock vs pop songs
Statues vs painPngs
Sea vs mountain vacaPon
Astrology
Machine Learning
Skills/knowledge Java
Excel
Novice in ML
Expert of astrology
NLP
What I am interested in
What I prefer
What I am interested in
What I prefer
What I know
Enjoyment
Learning
10. User profile - signals
Roberto Turrin
Personaliza0on Challenges in E-Learning
What I know
What I am interested in
Consuming a resource
What I am interested in
Educa4onStandard
The activities done by the user affect his skills.
In fact, as I study a change my knowledge, I learn more
about a topic, I increase my understanding, I enable
myself to learn something more complex on the same
topic. Since skills are part of my profile, I practically
change my profile. We can so say that use profile in
education really changes over time
Watching/discovering a new kind of movie
might modify my interests
11. User profile & User task
Roberto Turrin
Personaliza0on Challenges in E-Learning
Educa4on
User profile Java
Excel
Novice in ML
Expert of astrology
NLP
Learning Python
Preparing for a cerPficaPon
TesPng the level of knowledge
Mastering BBQ cooking
Becoming a data scienPst
User task
“Changing what I know”
What I want to achieve/know
What I know
13. Time evoluPon
Roberto Turrin
Personaliza0on Challenges in E-Learning
S3
BigQuery
0me
Learning Tes0ngLearning Learning
Recommender goal:
• “providing learning resources to make the user profile close to the user goal”
• “providing training resources to improve the confidence of user profile representa+on”
15. Heterogeneity - bundles and paths
Roberto Turrin
Personaliza0on Challenges in E-Learning
Video lectures Hands-on Quizzes
Learning paths Exams
16. User raPngs
Roberto Turrin
Personaliza0on Challenges in E-Learning
User ra+ngs not par+cularly useful for the recommender:
• They are rare. Most of user signals are implicit.
• They are more related to the quality of the resource than to the interest of the user
or to their uPlity for the user goal.
• They are more useful for the content producer than for the user as they represent a
feedback for the content.
In fact, there is a high correlaPon between the raPng mean and the number of
negaPve and posiPve comments.
17. Algorithms
User profile transparency is o^en a requirement:
• the user profile represents the current user skills
• the user is curious about “himself”
Roberto Turrin
Personaliza0on Challenges in E-Learning
Experiments with pure collabora0ve did not succeed
• not aligned with the user learning task
• a lot of new content
?
• Currently, a hybrid is being used
• Working on embedding learning tasks
through an ontology.
18. Other peculiariPes of on-line training: open points
Lack of a physical class:
• social features
• forum
• pair-tasks
Roberto Turrin
Personaliza0on Challenges in E-Learning
User recommendaPons
Time constraints:
• user learning pace
• user deadlines
• resource Pming
• resource Pme availability
Planning
19. Conclusions
Roberto Turrin
Personaliza0on Challenges in E-Learning
• Learning goals drive most of user consumpPons.
• User profile also represents skills.
• The main user goal can be translated into “changing my skills”, i.e., changing my
profile.
• Consequently,
• profile conPnuously changes over Pme.
• profile is something the user is interested into.
• Use carefully raPngs and collaboraPve filtering