Personal recommender systems for learners in lifelong learning networks
1. Personal Recommender
Systems for learners in
lifelong learning networks
Denny Abraham Cheriyan
Dhanyatha Manjunath
Roja Ennam
Shruthi Ramamurthy
2. Objectives
● Lifelong Learning and need for Recommendation
Systems in Learning Networks
● How requirements of recommendation systems are
different for learning communities
● Primer on recommendation techniques
● Knowledge Maps for Self Directed Learning and
Recommendation
3. Lifelong Learning
● Not just limited to childhood/school. Learn at work place,
personal growth
● Control - What,When, Where and How
● Self-Directed
● Freedom
● Learner Centric
● Demand-pull approach (Minds of Fire - John Seely Brown and
Richard P. Adler)
4. Why Recommendation in Learning networks
● Using recommendation to provide navigational support in a
learning network
● Need advice to decide on most suitable learning activities to
meet the learner’s individual learning goal (or Learner
group)
● Self-directed learners need an overview of the available
learning activities and must be able to determine which of
these would match their personal needs, preferences, prior
knowledge and current situation
5.
6.
7. Activity
Can we use commercial recommendation technique to a
learning management system ? Are there any specific
requirements of a LMS, you think that a recommender system
must consider?
8. The specific demands for learning
• the importance of the context of learning
• the inherent novelty of most learning activities
• the need for a learning strategy
• the need to take changes and learning processes into
account.
9. Personal Recommender systems for lifelong
learners(Requirements)
● Learning goal
● Prior Knowledge
● Preferences and Learner characteristics
● Learner grouping and stereotyping (e.g., study time, study
interests and motivation to learn)
● Ratings of the learning activities
● Historical information about the successful study behaviour
● Apply the learning strategies
10. Activity 1
Choose an online learning community. What type of
recommendations would you prefer to help achieve your
learning goals?
11. Data for Recommendation Systems
● Obtained from Data Analytics (Shum & Ferguson)
○ Social Network Analysis
○ Content Analysis
○ Discourse Analysis
○ Disposition Analysis
○ Context Analysis
● Learner Specific Analysis (Wise, Zhao, & Hausknecht)
○ Micro and Macro Level Analysis
13. Collaborative Filtering Techniques:
Collaborative Filtering Techniques use the collective behaviors
of all the learners in the Learning Network.
The following are some collaborative filtering techniques:
● User -based Collaborative Filtering.
● Item - based Collaborative Filtering.
14. User- based Collaborative Filtering
User-based techniques correlate users by mining their ratings
and then recommend new items that were preferred by similar
users.
Figure 1
16. Advantages of Collaborative
Filtering
Disadvantages of Collaborative
Filtering
● The information that is being
considered for
recommendation is domain
independent.
● It does not depend on
analysing the content
provided by the user.
● Cold start problem
● Can not handle new users
and new items
● Unpopular tastes are not
strongly supported(Long tail
in learning as discussed in
Minds of Fire is not possible)
17. Question
How do you recommend items to a user when he is new to the
community?
18. Prompt the user to rate certain movies before being able to provide personalized recommendation.
19. Content Based Recommendation Technique
● Maintains a user profile- a structured representation of user interests.
● Matches the descriptions of items with the attributes of the user profile to recommend
similar items.
● Results of analysis represents the user’s level of interest in that item.
● If user preferences are accurately reflected by a user profile, the recommendation is
advantageous.
● Useful in handling the ‘cold start’ problem as no behaviour data is needed.
20. Summary of Content based
Recommendation System
Advantages Disadvantages
Useful in solving cold start problem-
When a user is new to the community,
Overspecialization- items that are highly
correlated with user profile or interest are
only recommended.
Can map user needs to the items-
Attributes of the user profile are matched
with the attributes of the item.
Can work with information that can be
described as a set of attributes-
Learning content/Items must be classified
into categories.Needs Category modelling
and maintenance.Sensitive to changes to learner profile-
Change in the value of the attributes of
learner profile affect recommendation
results.
21. Usefulness of Recommendation Techniques in LN
Content based Recommendation
● Keeps the learner on track with his/her learning goals
● Helps the user to specialize in a domain.
● Useful in recommending items to a user, when he/she is new to the
community.
● Can map the characteristics of lifelong learners to that of learning
activities. For example: learning goal, prior knowledge, available study
time.
Collaborative Filtering
● Useful for Learner Networks which offer courses in different domains.
● Learning activities which are frequently positively rated and their sequence
can be identified as popular learning tracks.
● Could use learner information to group similar learners together in a
learning group.
22. Hybrid Recommendation Technique
● A combination of recommendation techniques constitutes a
hybrid recommendation or a recommendation strategy.
● Hybrid recommendation technique could provide the most
accurate recommendations by compensating for the
disadvantages of single techniques.
Example:
If no efficient information is available to carry out CF techniques, it
would switch to a CB technique.
● Uses historical information about users or items to decide
which specific recommendation technique provides the highest
accuracy.
29. Knowledge Maps and Khan Academy
● Each node in the graph represents a topic, with exercises
related to the topic. If the user masters the topic it turns
blue.
● Have to manually look at the entire concept map and
manually choose path
● The large number of options available will be overwhelming
30. Knowledge Maps and self directed learning
● Fits well with self-directed learning
● Not constrained by a course start date and end date
● Breakdown course into independent topics. Can change
path mid-way and still use whatever knowledge has been
accrued previously
● Can choose your own path based on goals
● Can discover new topics of interest (Poker and Probability,
Game Theory is related to both Economics and Artificial
Intelligence)
31. Knowledge Maps and Recommendation
● Knowledge maps can be overwhelming (Need an agent to
direct you)
● Users can be represented as paths in a network
● Can recommend paths, based on other users who have
mastered similar paths
● Ability to suggest paths based on goals (Search)
32. ● Domain knowledge required to construct knowledge maps
(prerequisites).
● Content curation, map construction will have to be done
manually
● Complexity in generalizing Knowledge Maps
Knowledge Maps - Disadvantages
33. Adaptive Intelligent Support Systems
● An adaptive support system improves learning and
collaboration
● Personalized and tailored to student needs
● Provides more relevant support as the number of support
instances increases
34. Recommending Users to Learner Groups
● Characteristics of the current learner could be taken into
account to allocate learners into groups.
● Stereotype recommendation technique can be used to
allocate learners into groups if no behavior data is available
● Learner grouping can focus on relevant learning
characteristics (eg - similarities in learning behavior, study
time, study interests, motivation to learn).
35. Factors that influence groups
● Personality characteristics (introvert/extrovert)
● Status (high/low status)
● Gender
● Race
● Level of conflict (too less or too much conflict)
● Lack of co-ordination
● Negative socio-emotional processes
36. Activity 2
From Piazza list out the information that can be captured to form learner groups.
How will you use this information to recommend who to collaborate with?
In Webbs paper, some of the factors that influence groups -
● Personality characteristics (introvert/extrovert)
● Status (high/low status)
● Gender
● Race
● Level of conflict (too less or too much conflict)
● Negative socio-emotional processes
37. References
● H. Drachsler, H.G.K. Hummel and R. Koper , ‘Personal recommender
systems for learners in lifelong learning networks: the requirements,
techniques and model’, Open University of the Netherlands
● Anderson, T. , ‘Toward Theory of Online Learning’ , Athabasca University,
(pp. 33–60)
● Ounnas, A., Davis, H. C., & Millard, D. E. (2009). A Framework for Semantic
Group Formation in Education. Educational Technology & Society, 12 (4),
43–55.
● John Seely Brown and Richard P. Adler, Minds of Fire
● Erin Walker & Nikol Rummel & Kenneth R. Koedinger, Adaptive Intelligent
Support to Improve Peer Tutoring in Algebra, International Artificial
Intelligence in Education Society 2013
● Simon Buckingham Shum and Rebecca Ferguson, Social Learning Analytics,
Technical Report KMI-11-01, June 2011
● Alyssa Friend Wise, Yuting Zhao, Simone Nicole Hausknecht, ‘A Pedagogical
Model for Intervention with Embedded and Extracted Analytics’, Simon Fraser
University
● Noreen M Webb, ‘Information Processing Approaches to Collaborative
Learning’
Editor's Notes
The demand-pull approach is based on providing students with access to rich (sometimes virtual) learning communities built around a practice.
It is passion-based learning, motivated by the student either wanting to become a member of a particular community of practice or just wanting to learn about, make, or perform something.
With the numerous number of options available, it might be overwhelming for the end-user to select learning activities.
finding their way through the possibilities offered
Match personal needs, preferences, prior knowledge and current situation
Deal with information about the learners (users)and learning activities (items) and would have to combine different levels of complexity
for the different learning situations the learners may be involved in.
distance education. - students rely on the information online - no face to face meetings. be aware of our
target group of learners.
Learning math is different from learning history
Learning strategy : easier courses before harder ones - preferred media and the characteristics of the learning content when designing instruction
various stages of learning , educational, psychological, social and cognitive science.
What learners want to learn
proficiency level of the learning activity should fit the proficiency level of the learner (prior knowledge).
preferences (e.g., preference for distance education or problem-based learning) for learning (learner characteristics).
Learners with the same learning goal or similar study time per week could benefit from the ratings received from more advanced learners.
historical information about the successful study behaviour of the more advanced learners in the same learning network
learning strategies derived from educational psychology research
‘go from simple to more complex tasks’ or ‘gradually decrease the amount of contact and direct guidance’.
Resources: Learning articles, Other users with domain expertise, Remedial Exercises. From an embedded and extracted perspective
Networz,smdsmdasdk Analysis
Scenario where there is no sufficient information to give the recommendations.This problem is because collaborative filtering depends a lot on the user behaviour from the past.
recommend courses
Central learner profiles where users can list courses they have taken from various MOOCS. Like linkedIn profile to solve the cold start problem..
Content based Recommendation is a technique which can be used in such scenario, which maintains user profile containing data to user profile.
It matches the preferences in user`s profile with the descriptions of the items, and
The use of a combination of
recommendation techniques constitutes a recommendation strategy
Bring it back to question 1
Recommendation strategies use domain-specific or historical information about users or
items to decide which specific recommendation technique provides the highest accuracy
for the current user
Capturing Learning goals - I'm assuming that based on your learning goal they'll decide what they do with your data. Drop out rates. I haven't finished a course on coursera yet. I generally watch videos on random topics in these courses
The audience of khan academy is targeted towards high school students
Have a lot of overlap between courses at ASU
Obtaining metadata information
Ontology modelling and maintenance is required
Can suggest various paths to goal. Based on time constraints, previous knowledge, etc. Analogous to Google Maps (Best Route, Fewer transfers, Less Walking)
Erins paper talks about how adaptive systems can assist tutors and foster collaboration
Current paper talks about grouping users based on similarity
Webbs paper talks about other factors that effect groups. Dhanyatha mentioned about Hybrid Collaboration. we could use a mix of recommendation techniques. We should strike the right balance(who defines what the right balance is) between these factors. As Dhanyatha mentioned about Hybrid Collaboration. We could use a mix of recommendation techniques
Where can information be captured - At signup, after sign up (user behavior), performance of previously assigned groups
Consider Piazza and the discussion activity we participate in. f the professor was not assigIning groups