Learning User Profiles from Tagging Data and Leveraging

Due to the high popularity of social bookmarking systems, a large amount of metadata is available. Aggregating the metadata belonging to one user results in an user profile similar to those often used in Information Filtering. This paper shows how to create user profiles from tagging data. We present the Add-A-Tag algorithm for profile construction which takes account of the structural and temporal nature of tagging data. In addition, we explore ways of leverag- ing these

user profiles. There are two main insights gained. Firstly, as we experienced in a small-scale user study, sim- ply being able to view aggregated information about past tagging behavior was considered useful. Secondly, the user profile can be used to guide the user’s navigation, that is, to provide the user with personalized access to information resources. Categories and Subject Descriptors H.3.3 [Information Storage and Retrieval]: Information Search and Retrieval Keywords tagging user profiles dynamics information filtering visuali- sation hci 1. INTRODUCTION Social bookmarking systems, such as del.icio.us [19], have been around for quite a while now. They provide interfaces for annotating bookmarks with free-text keywords. Their simplicity and their immediate usefulness for improved re- discovery of information have attracted a high number of users. All users’ annotated bookmarks are by default pub- licly accessible. Hence, an immense amount of metadata is available. This collaboratively created data is a valuable resource. Aggregations of it are provided to the user commu- nity. Several papers address analysis [9] or data mining [12] of tagging data. Most authors analyse the properties of the ∗This research has partly been funded by the Austrian Fed- eral Ministry for Education, Science, and Culture (bm:bwk), and the European Social Fund (ESF) under grant 31.963/46- VII/9/2002. Copyright is held by the author/owner(s). WWW2007, May 8–12, 2007, Banff, Canada. . metadata related to certain bookmarks and/or to certain tags. In this paper, we focus on those tags which have been employed by a certain user. We treat them as a continuous stream of information about a user’s interests, which can be used for creating a rich user profile. Aggregated information about a user’s bookmark collec- tion is usually represented as a tag cloud, in which all tags a user has employed so far are listed alphabetically and the font size of a tag is set according to how often it has been used so far. Our claim is that tag clouds fail to represent two important properties of a user’s bookmark collection. • Firstly, they do not represent the relationships be- tween the tags, which can be derived by using co- occurrence techniques. • Secondly, they do not consider that tagging data is time-based in their weighting of the relative impor- tance of a tag. Our aim is to learn user profiles from tagging data that in- clude those two properties. In addition to creating profiles, we need to present...

Website: publik.tuwien.ac.at | Filesize: -
No of Page(s): 8
Download Learning User Profiles from Tagging Data and Leveraging them for ....pdf