Article-level clustering

Just as we can map the structure of science by clustering at the journal level, we can map the scientific literature at a finer grain by clustering at the level of individual articles using article citation data such as that available in the Microsoft collection. Again we use the Map Equation (Rosvall and Bergstrom 2008 Proc Nat Acad Sci USA). Because of the time-directedness of the article-level citation network, we treat this as an undirected graph. Recent developments (Bergstrom and Rosvall, unpublished) in clustering time-directed networks may further improve on the categorization presented here

Because the number of categories found in this way will be large, it is useful to turn to semantic labeling approaches to automatically assign names to categories and facilitate user navigation.

Using the article-level category browser

Article titles are provided on the left-hand side. The Eigenfactor column provides the relative ranking of the articles within each category, such that the most important articles within each cluster are listed at the top of the cluster. (Alternatively, articles within each cluster could be ordered in terms of their centrality to that individual cluster; these would be "most-representative" instead of "most influential". Either of these might provide valuable recommendations to a user reading another paper within any given cluster.)