Latent subject-centered modeling of collaborative tagging: An application in social search

Collaborative tagging or social bookmarking is a main component of Web 2.0 systems and has been widely recognized as one of the key technologies underpinning next-generation knowledge management platforms. In this article, we propose a subject-centered model of collaborative tagging to account for the ternary cooccurrences involving users, items, and tags in such systems. Extending the well-established probabilistic latent semantic analysis theory for knowledge representation, our model maps the user, item, and tag entities into a common latent subject space that captures the "wisdom of the crowd" resulted from the collaborative tagging process. To put this model into action, we have developed a novel way to estimate the probabilistic subject- centeredmodel approximately in a highly efficient manner taking advantage of amatrix factorization method. Our empirical evaluation shows that our proposed approach delivers substantial performance improvement on the knowledge resource recommendation task over the state-of-the-art standard and tag-aware resource recommendation algorithms.