Direct Citation Clustering of Science

All three build networks among publications but use different relations. Co-citation links two papers when later work cites them together; bibliographic coupling links two papers when they share references; direct citation links a citing paper to the paper it cites, using the citation itself as the edge. Direct citation is the most explicit and scales well, but because it depends on citations actually existing between papers, very recent work can be weakly connected — which is why coupling, formed at publication, often represents the newest front sooner.

The resolution limit is a known weakness of standard modularity optimization: in large networks it tends to merge or overlook clusters below a certain size, even when they are genuine communities. For a direct-citation network of millions of papers, this would hide many real research areas. Waltman and van Eck address it with a quality function carrying an explicit resolution parameter and suitable normalization, and with scalable algorithms, so that small but coherent specialties can be recovered alongside large fields by tuning the resolution.

Journal-based classification assigns every paper in a journal to the journal's field, which misplaces interdisciplinary papers and any work that sits outside its journal's typical scope. Publication-level direct citation clustering instead assigns each paper according to its own citation relations, so a chemistry paper in a general-science journal lands in chemistry, and an interdisciplinary paper joins the area it is actually connected to. This yields a more accurate and finer-grained map, which is the central advantage Waltman and van Eck's methodology was designed to deliver.