Evolutionary taxonomy construction from dynamic tag space

Abstract

Collaborative tagging becomes a common feature of current web sites, facilitating ordinary users to annotate and represent online resources. The large collection of tags and their relationships form a tag space. In this kind of tag space, the popularity and correlation amongst tags capture the current social interests. Tags are freely chosen keywords and difficult to organize. As a hierarchical concept structure to represent the subsumption relationships, automatically extracted taxonomies become a viable method to manage collaborative tags. However, tags change over time, and it is also imperative to incorporate the temporal tag evolution into the extracted taxonomies. In this paper, we formalize the problem of evolutionary taxonomy generation over a large collection of tags. A line of taxonomies are generated to reflect the temporal changes of underlying tag space. The proposed evolutionary taxonomy framework consists of two novel contributions. First, we develop a context-aware edge selection algorithm for taxonomy extraction. This method is built on seminal association-rule mining algorithm. Second, we propose several strategies for evolutionary taxonomy fusion, which smooths the newly generated taxonomy with prior ones. We conduct an extensive performance study using a large real-life web page tagging dataset (i.e., Del.ici.ous). The empirical results clearly verify the effectiveness and efficiency of the proposed approach.