Discovery of Social Navigation Patterns and Friendship Relations in Social Bookmarking Web Site

1
Discovery of Social Navigation Patterns and
Friendship Relations in Social Bookmarking Web
Site
Department of Computer Science
EngineeringCollege of Engineering
Feilong Chen, Pan-Ning Tan

• Social bookmarking tools enable users to save
  URLs for future references and share their
  bookmarks with other users.
• This work is a case study on the application of
  data mining to a popular social networking web
  site called del.icio.us.
• Goal is to show that analysis of the bookmarks
  and tags may reveal useful insights into
• Social navigation behavior of Web users
• What makes a web page popular?
• What makes a user (bookmarker) popular?
• Who make friends with whom?
• Friendship relations between users

• What makes a web page popular?
• Potential reasons
• Bookmarked by popular users
• Contains popular tags
• Has high PageRank score
• Early bookmarkers are authoritative users
• Authoritativeness is measured by fraction of
  popular bookmarks in which the user is an early
  bookmarker
• Experiments showed that popularity of a
  bookmarker is not correlated with popularity of a
  web page
• Using the remaining features to build a decision
  tree classifier, we were able to predict popular
  bookmarks at 63 accuracy.

3. Who make friends with whom? The problem Given
a set of users, their bookmarking activity and
tagging activity, we want to predict their friend
relationship. Algorithm Kernel Alignment Given
two kernels K1 and K2, their alignment is
measured by

• Del.icio.us Social Bookmarking
• Users save and share bookmarks
• Users assign tags to bookmarks
• A user can add others to his own network
• He becomes a fan of those users

• Prediction algorithm
• Compute pairwise similarity between users based
  on bookmark and user tag data
• Align the similarity matrices to a subset of
  users friendship relation (training data)
• Use the values of the aligned kernel matrices as
  features for SVM classifier

• 2. What makes a user popular?
• Potential factors
• Number of bookmarks saved
• Number of times being an early bookmarker
• Popularity of users frequently used tags
• Experiments have shown that the first two factors
  are positively correlated with a users
  popularity (number of fans).
• A decision tree built on the two features can
  achieve 61 accuracy in predicting popular users.

Date goes here