Figure 1.
Bibliographic Coupling Network Communities in the Complete Corpus.
Panel A presents the full bibliographic coupling network, edge-reduction is based on papers with weighted similarity scores two standard deviations above the median similarity among non-isolates in the network. Node color represents each paper's identified bibliographic coupling community using the Newman-Girvan algorithm [26]. Panels B and C present the same analyses limited only to publications from AIDS and JAIDS respectively. Panel D show the correspondence between communities and the broad “discipline” like labels applied to all published articles beginning in 1998. Color represents whether a label is over (blue) or under (red) represented in a given community according permutation-based residuals.
Figure 2.
Community-Topic (lack of) Correspondence.
This mosaic plot shows those topics that are overrepresented present in more than one network community (top 11), or are not consolidated in any community (bottom 2). The topics are derived via LDA (see Supplementary Information) and the communities are those represented in Fig. 1. Color represents over (blue) and under (red) representation of a topic in a given community according to permutation-based residuals.
Figure 3.
Temporal change in modularity, 1988–2008.
Constructed networks comprise all articles published in a 4-year moving window (with labeled year indicating the ending year of that window). For each temporal slice, community detection is applied, and the summary modularity index is presented. The 1998 dip follows the introduction of “discipline” like labels for on all published articles.
Figure 4.
Alluvial Flow Diagram w/“Discipline” Like Labels.
This figure presents the evolution of clusters within 5-year moving windows (reduced to include only clusters containing at least 50 papers). The color corresponds to clusters in which the broad “discipline”-like labels are over-represented in a given community (yellow = Social/Epidemiological, blue = Basic, red = Clinical).