How bibliometric networks are built
Every network starts from a document-by-item table: which keywords, references, authors or countries appear in each document. Two items are linked when they appear in the same documents, and the link weight is the number of documents they share. Three choices shape the map:
- Number of items and thresholds. Keep the 50–100 most frequent items for a readable map, and raise the minimum link weight to remove chance connections.
- Normalisation. Raw counts favour frequent items. Association strength, the ratio of observed to expected co-occurrences, is the recommended default (van Eck & Waltman, 2009). Salton’s cosine, Jaccard, inclusion and equivalence indices are also available.
- Clustering. The Louvain algorithm (Blondel et al., 2008) groups items so that links inside clusters are stronger than expected by chance (modularity). Increase the resolution for more, smaller clusters.
Node centrality helps interpretation: degree counts links, betweenness identifies bridges between clusters, and PageRank rewards links from important nodes.
Keyword co-occurrence network
Keywords that are used together in the same documents form clusters, and each cluster is a theme. Use author keywords for the authors’ own framing, Keywords Plus or index keywords for broader coverage, or title and abstract terms when keywords are missing. Merge synonyms first. The overlay view colours terms by average publication year, so recent topics stand out; the density view shows where research concentrates.
Co-citation analysis
Two references are co-cited when a later document cites both (Small, 1973). Frequently co-cited references belong to the same school of thought, so co-citation clusters reveal the intellectual base of the field. Co-citation can also be run on cited first authors or cited sources. Because it depends on future citing behaviour, it is best for established literatures.
Bibliographic coupling
Two documents are coupled when they cite the same references (Kessler, 1963). Coupling is fixed once a paper is published and works for recent papers, so it maps current research fronts. It can be aggregated to authors, sources, institutions or countries. The clustering-by-coupling map plots coupling clusters by centrality and impact (their documents’ normalised local citations), and labels each with its most frequent keywords.
Collaboration networks
Co-authorship networks link authors, institutions or countries that publish together. Clusters show research groups or regional blocs, and high-betweenness nodes are brokers. The world collaboration map draws international co-authorship as arcs. The corresponding author’s country table separates single-country (SCP) from multiple-country (MCP) publications.
Thematic map
Themes are clusters of the keyword co-occurrence network. Each theme is placed by two measures (Callon et al., 1991; Cobo et al., 2011):
- Centrality (x-axis): the strength of a theme’s links to other themes, or how important it is to the field.
- Density (y-axis): the strength of the links among its own keywords, or how developed it is.
| Quadrant | Centrality | Density | Interpretation |
|---|---|---|---|
| Motor themes | High | High | Well developed and central: the field’s driving topics |
| Niche themes | Low | High | Well developed but specialised and isolated |
| Emerging or declining themes | Low | Low | Marginal: check the trend over time to decide which |
| Basic themes | High | Low | General, transversal topics that are not yet well developed |
Thematic evolution
The collection is split into periods at the cut years you choose, a thematic map is built for each period, and themes in consecutive periods are linked when they share keywords. The inclusion index (shared keywords divided by the size of the smaller theme) measures the strength of each link. The result is a Sankey diagram of themes splitting, merging, appearing and disappearing. Choose cut years at meaningful events and keep enough documents in each period.
Factorial analysis
Multiple correspondence analysis (MCA) places keywords on a two-dimensional map so that keywords used by similar documents sit close together. Ward hierarchical clustering then groups them into topics, and the topic dendrogram shows how the topics merge. Report the share of inertia explained by each dimension, and interpret clusters rather than single positions.
Roots: historiograph and reference spectroscopy
The historiograph (Garfield, 2004) places the most locally cited documents on a timeline with arrows for direct citations between them, showing how the key ideas developed. Reference publication year spectroscopy (RPYS; Marx et al., 2014) counts cited references by their publication year; years that stand out above the 5-year median point to seminal works.
Fronts: trend topics and burst detection
Trend topics show, for each term, the interquartile range of the publication years of the documents that use it; terms with recent medians are new research fronts. Burst detection (Kleinberg, 2003), popularised by CiteSpace, finds terms whose use rose sharply for a period, and reports the start, end and strength of each burst.
References
- Blondel, V. D., Guillaume, J.-L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics, 2008(10), P10008.
- Callon, M., Courtial, J. P., & Laville, F. (1991). Co-word analysis as a tool for describing the network of interactions between basic and technological research. Scientometrics, 22(1), 155–205.
- Cobo, M. J., López-Herrera, A. G., Herrera-Viedma, E., & Herrera, F. (2011). An approach for detecting, quantifying, and visualizing the evolution of a research field. Journal of Informetrics, 5(1), 146–166.
- Garfield, E. (2004). Historiographic mapping of knowledge domains literature. Journal of Information Science, 30(2), 119–145.
- Kessler, M. M. (1963). Bibliographic coupling between scientific papers. American Documentation, 14(1), 10–25.
- Kleinberg, J. (2003). Bursty and hierarchical structure in streams. Data Mining and Knowledge Discovery, 7(4), 373–397.
- Marx, W., Bornmann, L., Barth, A., & Leydesdorff, L. (2014). Detecting the historical roots of research fields by reference publication year spectroscopy (RPYS). Journal of the Association for Information Science and Technology, 65(4), 751–764.
- Small, H. (1973). Co-citation in the scientific literature: A new measure of the relationship between two documents. Journal of the American Society for Information Science, 24(4), 265–269.
- van Eck, N. J., & Waltman, L. (2009). How to normalize cooccurrence data? An analysis of some well-known similarity measures. Journal of the American Society for Information Science and Technology, 60(8), 1635–1651.