Tag navigation

ABSTRACTThe amount of information available on the world wide web keeps growing at an exponential pace. Social tagging is a feature of various online social networks to organize information elements by letting people label these with free-form text, called tags. The graph created by this process is often called a folksonomy and comprises the association between people, tags and documents. Tagging is now used to organize web pages, pictures, videos, music, books, academic publications, etc. The current ways of navigating folksonomies are limited. In most web portals, "search" is the main feature which uses tags. When browsing tags, most systems give a few related tags to the clicked tag, none enables the user to get related tags to multiple clicked tags at the same time. A popular tag cloud displays links to the most popular tags in the folksonomy with a font size that depends on their popularity. Popular tag clouds and related tags can enable tag-based navigation. Enabling navigation through related tag clouds to multiple clicked tags in an efficient and scalable manner is a hard problem. We propose a bayesian approach to the problem of generating related tag clouds for navigation by using social network information and probabilistic models of people's tagging behaviors. We propose two new models to generate tag clouds based on popularity, tag co-occurrence and social relationships. The models are implemented in a prototype application to navigate empirical data from "last.fm", an online social network for music. We give an evaluation plan to compare the models regarding searchability through user evaluations.