Title: Design and Evaluation of an Adaptive Incentive Mechanism for Sustained Educational Communities
1Design and Evaluation of an Adaptive Incentive
Mechanism for Sustained Educational Communities
- Ran Cheng and Julita Vassileva
- User Modeling User-Adapted Interaction
- Presented by Rosta Farzan
- PAWS Group Meeting
- April 13, 2007
2Problem
Insufficient user participation in online
communities
Provide incentives to encourage participation
Too much participation especially of low-quality
3Goal
- Proposing an incentive mechanism regulating
quantity and quality of user contribution and
ensuring a sustainable user participation
4Outline
- Related work
- Collaborative quality evaluation mechanism
- Brief introduction to Comtella
- Proposed mechanism
- Implementation of proposed mechanism in Comtella
- Evaluation
- Discussion
5Collaborative Quality Evaluation Mechanism
- Active and sustained user participation while
avinding information overload requires quality
control - Decentralized moderation
- Real world example
- Measuring the quality of journals or papers by
counting the time they were cited - Online communities
- Counting the number of clicks on each item
- Problem
- Not always a click means a positive attitude
- Explicit user rating
- E.g. rating process in Slashdot
- Problem rich get richer
- Unfair score for items with insufficient
attention - Lower initial rating
- Contributed late in discussion
6Requirement of Incentives
- Time based community need
- New contributions in the early period of
discussion - Ratings of the contribution when many
contributions are collected - Different users have different contribution
patterns - Encourage higher participation for users who
contributed few high-quality resources - Inhibit contributions from users who contributed
many low-quality resources - Overall community need
7Comtella
- Developed at University of Saskatchewan
- Online community for sharing URLs of
class-related web-resources (bookmarks)
8Proposed Incentive Mechanism
- Mechanism encouraging users to rate resources
- Adaptive reward mechanism
9Encouraging Users to Rate
- All users can rate others contribution
- Each user receives a limited number of rating
points to give out - Users with higher level membership receives more
points - Initial rating of zero independent of level of
resource provider - Ensuring all contributions have equal chance to
be read and rated - Summative rating
- Incentive Virtual currency (c-points)
- Limited initial c-points
- Awarding users for rating resources
- Depending on users reputation
- Can be invested to promote the initial visibility
10Sorted by c-points
Summative Ratings
Report duplicate, broken, or special permission
links
11Adaptive Reward Mechanism
- Hierarchical membership
- Adapt rewards for different forms of
participation - Quality of users participation so far
- Community need
- Personalized motivational message
- Stating specific performance goals
- Calculating adaptive rewards
- Community Model
- Individual Model
12Community Model
- Expected of total contributions
- Set by community admin
- Community Reward Factor
- Usefulness of new resources
13Individual Model
- Contribution reputation
- Sharing
- Average summative rating of all shared resources
- Rating
- Quality Difference between the specific rating
and the average of all ratings the resource gets
eventually - Smaller difference ? Higher quality
- Current membership level
14Individual Model
- Expected of resource contribution
- Reward Factor
15Adaptive Reward Mechanism
16Implementation of proposed mechanism in Comtella
17Evaluation
- Questions to be answered
- Will the users in the test group rate articles
more actively? - How well will the summative ratings reflect the
real quality of the articles? - Will the users tend to share resources earlier in
the week? - Will the actual number of contributions be close
to the desired one? - Will the users share the number of articles that
is expected from them? - Will the users contribute a higher percentage of
high-quality articles? - Will there be information overload?
18Participant
- Class on Ethics and Information Technology
- Share web-articles related to the topic of each
week - 31 4th year undergraduate students
- Test group 15
- Control group 16
- Randomly assigned while controlling gender and
nationality
19Result
- Users in the test group were more active in
rating articles - The articles with higher ratings were more likely
to be chosen by users to summarize - The users in the test group were more satisfied
with the summative ratings received by their
articles - The users in the test group tended to share
resources earlier in the week - No big difference between total number of shared
articles across the two groups - In both groups, the users attitude towards the
quality of the articles were generally neutral - No information overload problem the overall
contribution did not exceed the community need - The users in the test group were more active in
terms of logging on the system and reading
articles sustainability
20Discussion/Limitations
- Choosing the Parameters for the mechanism
- Expected sum of contributions for each topic
- Threshold for reputation values
- Community reward function
- Narrow scale of rating (1 or -1)
- Less cognitive load
- No option to rate different aspect of a paper
- Popularity of topic, originality, interestingness
- Measure of quality
- Average rating means average taste rules
- In educational context tends to be superficial,
easy to read articles
21Discussion/Connection to Our Work
- Trying personalized motivational message in
CourseAgent or CoPE - Good evaluation questions
- Some people are easy to be motivated by glory
and recognition - Is there any cognitive tool to measure this?
- Compare Comtella with CoPE
- CoPE has the option to write summary and read
summary written by others - More personal benefit