Personalizing Navigation in Folksonomies Using Hierarchical Tag Clustering

Abstract

The popularity of collaborative tagging, otherwise known as “folksonomies”, emanate from the flexibility they afford users in navigating large information spaces for resources, tags, or other users, unencumbered by a pre-defined navigational or conceptual hierarchy. Despite its advantages, social tagging also increases user overhead in search and navigation: users are free to apply any tag they wish to a resource, often resulting in a large number of tags that are redundant, ambiguous, or idiosyncratic. Data mining techniques such as clustering provide a means to overcome this problem by learning aggregate user models, and thus reducing noise. In this paper we propose a method to personalize search and navigation based on unsupervised hierarchical agglomerative tag clustering. Given a user profile, represented as a vector of tags, the learned tag clusters provide the nexus between the user and those resources that correspond more closely to the user’s intent. We validate this assertion through extensive evaluation of the proposed algorithm using data from a real collaborative tagging Web site.