On behalf of the Knowledge Management Department and the competence centers CCSW and C4, we invite you to the following talk by Andreas Hotho, Universität Kassel:
DATE: Friday, June 29th, 2007
TIME: 15:00 - about 16:15
ROOM: 2.15, DFKI Kaiserslautern, Trippstadter Straße 122
TITLE:
Data Mining in Social Bookmarking Tools
ABSTRACT:
Social bookmark tools are rapidly emerging on the Web. In such systems, users are setting up lightweight conceptual structures called folksonomies. The reason for their immediate success is the fact that no specific skills are needed for participating and direct benefits are provided. BibSonomy is a folksonomy system developed and maintained at the university of Kassel. It gives you the opportunity to share not only bookmarks, but also publications in an integrated way. BibSonomy offers a large variety of functionalities which researchers often need in their daily work and acts as a playground for our research in folksonomies.
Andreas will start the talk with a short demonstration of the BibSonomy system and present then several approach to analyze folksonomies.
- Starting from the properties of the system we derive a formal folksonomy model and analyze the underlying graph structure along two real world datasets.
- As a next step we try to extract tag relations and to identify user communities. By apply association rule mining on an adapted graph structure we are able to identify relations between tags and resources. While the tag relations can be used to structure the tag cloud of a user, relations between resources are the basis for identifying communities within such systems. To come up with a topic description of the communities we compute labels by applying our FolkRank algorithm which takes advantage of the underlying hypergraph structure. In this way also the core users are identified.
- The last topic of the talk investigates the problem of tag recommendations. FolkRank is used again to recommend personalized tags for a new resource. Experimental results on three real world datasets show that FolkRank is able to outperform collaborative filtering approaches by 10%.
SPEAKER:
Andreas Hotho
http://www.kde.cs.uni-kassel.de/hotho
hotho@cs.uni-kassel.de
Universität Kassel
Fachgebiet Wissensverarbeitung
If you want to meet with the speaker (who will be in KL the whole day), please contact Michael Sintek.
Andreas Dengel (head of Knowledge Management Department)
Michael Sintek and Paul Buitelaar (co-heads of CCSW)
Stephan Baumann (head of C4)