Document Clustering, Visualization, and Retrieval via Link Mining

Clustering for document retrieval has traditionally been done through word-based similarity. But this approach suffers from the ambiguity problems inherent in natural languages. Language-based processing can be augmented by analysis of links among document sets, i.e. hypertext Web links or literature citations. Indeed, early workers in information science recognized the shortcomings with word-based document processing. This led to the introduction of document processing based on literature citations [6]. An important development was the notion of co-citation [13], in which a document pair is associated by being jointly cited (or co-cited) by other documents. In general, clustering based on co-citation as a similarity measure is known to correspond well to document semantics. Spurred by the popularity of the Web, more recent approaches have been developed for analyzing hyperlinks, though primarily for search engine page ranking [12, 8].