1. Personalized Retrieval in
Social Bookmarking
Scott Bateman, University of Saskatchewan
Michael Muller, Center for Social Software, IBM Research
Jill Freyne, CLARITY, University College Dublin
7. finding bookmarks
• filters: pivot browsing or typed tag filter
• 59% of filters lead to refinding a bookmark
– refinding: selecting a bookmark that has been
previously visited
– more refinding than discovery
8. bookmark refinding scenario
I need to find that
news article I saw in
Dogear about
collaboration and
social networking in
the workplace.
John
17. new ordering options needed
• list orderings don’t necessarily reflect what is
relevant to a user’s purpose
• move relevant bookmarks to the top of the list
– reduce user effort
18. evaluation of new metrics
• using system logs, identified all query sessions
in a 6 month period where users filtered lists
and selected a bookmark (a target)
– used all session whether refinding or not
• recreated query sessions comparing original
date-based ordering versus new ordering
– positions in result lists for target (rank)
– number of results lists where target was visible
19. wisdom of the crowd
• our initial attempts:
– access histories of all users
– access histories of automatically created
groups – based on cosine sim. of accesses,
tags, or bookmarks
24. we also found…
• improved result orderings on all filter types
(by tag, user, or user and tag)
• worked well on profiles of other users ->
suggests refinding?
25. summary
• Personalized orderings based on access
histories provide a simple metric for re-
ordering bookmarks
– improved position in list
– presented after fewer refinement steps
26. future work
• is there a way to incorporate group interaction
histories?