Social Systems: Can We Do More Than Just Poke Friends
Download
Report
Transcript Social Systems: Can We Do More Than Just Poke Friends
XUE Tianfan
The Chinese University of Hong Kong
This work is by Georgia Koutrika, published on CIDR'09
All the figures & tables in these slides are from that paper
Outline
Motivation
CourseRank
Unique features
Lessons Learnt so Far
Interaction with rich data
Conclusion
Motivation – CourseRank
Motivation
Social Web Site
FaceBook, del.icio.us, Y! Answer, Flickr, MySpace
Great success
Is it interesting for research community?
Are there any interesting chanllenges to researchers?
Can we do more than just poke friends?
Motivation
Social Web Site V.S. Traditional Open Web V.S. Database
Social Web Site
- mostly unstructured
- Centrally stored
- Users-to-Users Access Control
Traditional Open Web
- Unstructured
- highly distributed in storage
- Many provider and consumers without access control
Database
- Structured
- Centrally stored
- 1 provider, many consumers
Motivation
Social Web Site V.S. Traditional Open Web V.S. Database
Motivation
Research topics in database
Research topics in Web search
What is important for social website
What is most effective way for users to interact?
What can be shared among the users?
What information can be trusted?
How users to visualize and interact with information?
How users interact with other users?
How system evolve over time?
CourseRank
CourseRank
An educational social site where Stanford students can
explore course offerings and plan their academic program
Describe the insight of CourseRank in this paper
CourseRank
What CourseRank can do
Search for courses
Rank courses
Requirement check
Feedback to faculties
etc.
CourseRank
Unique features
Hybrid system – database + social system
Rich data
New tools – plannar, requirement checker, CourseCloud, etc.
Site Control
Closed Community & Restricted Access
Constituents
Lessons Learnt so Far
Lessons Learnt so Far
Meaningful Incentives
-
-
Yahoo! Answers:
Best answer – 10 points, vote for best answer – 1 point
CourseRank:
Different tools: planner, Q&A forum seeds
Interaction for Constituents
- Department Requirement
both useful for staff and students
Lessons Learnt so Far
Lessons Learnt so Far
Meaningful Incentives
-
-
Yahoo! Answers:
Best answer – 10 points, vote for best answer – 1 point
CourseRank:
Different tools: planner, Q&A forum seeds
Interaction for Constituents
- Department Requirement
both useful for staff and students
Lessons Learnt so Far
Lessons Learnt so Far
The power of a closed community
-
-
Block spammers and malicious users
User are more willing to contribute
Example: group forum, department forum, school forum, public
forum
It’s the Data, Stupid
-
-
External data
Hard to be shared data
Lessons Learnt so Far
Lessons Learnt so Far
Privacy can be “shared”
-
The course planned to be taken of a student -> closed
community
Closed Loop Feedback
-
Build by stanford students theirself, quickly get feedback
Beyond CourseRank: The Corporate Social Site
-
Example: Inner forum of a company
Can corporate social site learn something from CourseRank?
Interaction with Rich Data
Rich data
A student want to take a course:
Course name&description, user’s profile(major, class,
grade), course interrelationships, user’s comments, etc.
Problem of typical search engines
a student want something related to Greece
Search “Greece” -> no result
Search “Greek, science” -> got the course “history of science”
Search engine does not provide user specific result
“Java” is a good course, but not fit for non-engineering students
Interaction with Rich Data
Data Clouds
A data cloud is a tag cloud, where the “tags” are the most
representative or significant words found in the results of a
keyword search over the database.
Example:
“American” -> “Latin American”, “Indians”, and “politics”.
“American”: 1160 courses
“Latin American”: 123 courses
Challenge:
Multiple relation: tags does not only appear in course name and
description. For example, “java”.
How to rank the result
How to dynamically and efficiently update cloud
Interaction with Rich Data
Data Clouds
Interaction with Rich Data
Flexible Recommendation (FlexRecs)
Why
Provide recommendation is not easy considering multiple
connections. It need to be manually adjusted.
Previous recommendation algorithm is fixed
Interaction with Rich Data
Flexible Recommendation Example
Relations:
Simple reconmmendation example
Interaction with Rich Data
Flexible Recommendation Example
Complicated reconmmendation example
: recommend
: Expand
: Select
: Connect
Conclusion
Social sites:
A closed, well defined community
Provide rich data
Not simply for sharing links and networkings
Two mining tools
Data clouds
FlexRecs
Q&A