Treelicious: a System for Semantically Navigating Tagged Web Pages

Treelicious: a System for Semantically Navigating Tagged Web Pages Matt Mullins Department of Computer Science Western Washington University Bellingham, WA USA mtt.mllns@gmail.com Perry Fizzano Department of Computer Science Western Washington University Bellingham, WA USA perry.fizzano@wwu.edu Abstract—Collaborative tagging has emerged as a popular and effective method for organizing and describing pages on the Web. We present Treelicious, a system that allows hierarchical navigation of tagged web pages. Our system enriches the navigational capabilities of standard tagging systems, which typically exploit only popularity and co-occurrence data. We describe a prototype that leverages the Wikipedia category structure to allow a user to semantically navigate pages from the Delicious social bookmarking service. In our system a user can perform an ordinary keyword search and browse relevant pages but is also given the ability to broaden the search to more general topics and narrow it to more specific topics. We show that Treelicious indeed provides an intuitive framework that allows for improved and effective discovery of knowledge. Keywords-collaborative tagging; folksonomy; semantic web; social bookmarking; Wikipedia; Delicious I. INTRODUCTION Collaborative tagging has emerged as a popular and effective method for organizing and describing pages on the Web. There exist many different sites in different domains that use the application of free-form keywords as a method for organizing and searching their content. To name just a few: CiteULike for managing and discovering scholarly ref- erences, LibraryThing for cataloging and sharing literature, Etsy for buying and selling handmade items, and Delicious1 for organizing and sharing bookmarks. Tagging becomes especially useful to describe non-text media like photos on Flickr and videos on YouTube. These sites have embraced tagging as an effective and low-cost way of describing and organizing their content. On Delicious, one of the most popular social bookmarking sites, users annotate pages with tags, usually for the selfish reason of personal organization. Yet when this is done by many individuals, collectively rich and accurate descriptions of what these resources mean to humans materializes. Even though users are using tags primarily to help themselves retrieve the page later, 62% of the tags in Delicious end up identifying descriptive facts about the web resource—tags useful beyond personal 1http://delicious.com/ organization [1]. This user-generated classification structure has come to be known as a “folksonomy2”. Yet these folksonomies are lacking in several ways. First, they’re flat. There is no explicit hierarchy, synonymy, or relation information present—only simple co-occurrence data. Second, they’re ambiguous. This is the classic problem of using words with multiple meanings and no explicit disambiguation information. Given this lack of semantics there are only a handful of ways we can present sets of tags to the user. A common method is to use a tag “cloud” with more popular tags in the cloud indicated by a larger font size. Another method is to start with a search tag and present related tags based on which tags the search tag co- occurs with in tagged content. This co-occurrence data can also be used to group related tags using clustering techniques as is done in Flickr. Though all of these methods are helpful in some way, ultimately, they fail to show the semantic relationships among tags [2]. As a result, it is hard for a user to put their search into perspective. Figure 1 shows an example of the related tags produced from a search for “acm” on Delicious. Figure 1. A Delicious search for “acm”’ yields these “related” tags. Their relation is based solely on co-occurrence information. 2http://vanderwal.net/folksonomy.html arXiv:1102.1111v1 [cs.IR] 5 Feb 2011 This lack of structure resulting from the use of free- form tags is not encountered with more classic systems of classification like hierarchical taxonomies and library classifications. The categories in these systems are well- defined and placed in a strict hierarchy. Each subcategory can have only one parent category of which it is a member. Such a structure results in clear semantic “broader than” and “narrower than” relationships among concepts. But the strict- ness inherent in these classic systems presents disadvantages. They require expert catalogers, authoritative sources of judg- ment, and users educated about the categories [3]. It also takes work to keep them from becoming outdated as new categories are formed and old ones are restructured (e.g. “the Soviet Union” being reclassified as a “Former country”). Commenting on the restriction that each class have only one parent, Voss [4] observes that “Hierarchy seems to have a strict semantic that does not fit to the vagueness of the world. In practice there are always several ways to classify an object .. . If one uses polyhierarchy like in a thesa

This content is AI-processed based on open access ArXiv data.