Science Mapping

Citation-based maps (co-citation, bibliographic coupling) show intellectual influence and recognized similarity as determined by how papers are cited. These maps are robust and historically validated but require time for citation accumulation. Keyword-based maps (co-occurrence) show current research interests and terminology; they detect emerging work immediately but are more subject to noise and transient trends. Best practice: create both and compare. Citation maps show established structure; keyword maps show current focus. Discrepancies reveal paradigm shifts (researchers shifting to new terminology or methods).

Cluster size: larger clusters are more active subfields. Cluster density: tightly clustered nodes indicate cohesive communities; loosely clustered nodes indicate emerging or diverse subtopics. Cluster isolation: isolated clusters indicate specialized subfields with little cross-reference to other areas. Cluster labels: examine papers in each cluster to assign meaningful labels; avoid over-interpreting cluster positions as conceptual boundaries. Always validate clusters against domain expertise—not all statistically detected clusters are scientifically meaningful.

Force-directed layouts (spring embedders) are most common and interpretable; they place similar nodes close together. For large networks (>10K nodes), use algorithms optimized for speed (e.g., Fruchterman-Reingold with approximations). For very large networks (>100K nodes), use graph partitioning methods (clustering first, then layout each cluster separately). Experiment: try multiple algorithms (Kamada-Kawai, Yifan Hu) and compare—different algorithms may reveal different patterns. For published maps, be explicit about algorithm choice and parameters.

Create time-sliced maps (e.g., 2-year windows). A new cluster appearing in the most recent window (not present in prior windows) is an emerging field. Alternatively, track keywords with rising frequency (bursts) or papers with rapid citation accumulation. Emerging fields often appear as small, loosely connected clusters at the periphery of maps; they strengthen (densify) and integrate with the central landscape as they mature.