Why researchers choose VOSviewer
VOSviewer was developed at the Centre for Science and Technology Studies (CWTS), Leiden University, and introduced by van Eck and Waltman (2010) as a program for constructing and viewing bibliometric maps. Its VOS mapping technique places strongly related items close together, and an integrated clustering algorithm colours groups of densely connected items. From a single Scopus or Web of Science export you can build co-authorship, keyword co-occurrence, citation, co-citation and bibliographic coupling networks.
The software is popular because it handles large networks gracefully and produces clear, attractive maps. The catch is that output quality depends heavily on input quality and parameter choices. Without a thesaurus, a single concept such as machine learning may appear as several separate nodes, and a single research stream may be scattered across three clusters for purely lexical reasons.
It also helps to know what VOSviewer is not designed for. It does not calculate descriptive statistics such as annual growth rates, h-index rankings or Lotka distributions, and it has no built-in way to compare themes across time slices beyond the overlay colour scale. Where a project needs those outputs, we pair VOSviewer with Bibliometrix in R so that each tool does what it does best. VOSviewer does, however, offer a useful text-mining mode that extracts noun phrases from titles and abstracts, which we use when author keywords are sparse or inconsistent in older records.
Common VOSviewer problems we fix
Many clients come to us after a first attempt of their own. The issues are usually some mix of the following:
- Hairball maps: thousands of nodes with overlapping labels that become unreadable when printed in a journal column.
- Odd clusters: unrelated concepts grouped together, or one coherent theme fragmented into many small clusters.
- Reviewer questions: requests to justify the occurrence threshold, explain the counting method or rerun with fractional counting.
- Merged data: Scopus and Web of Science files use different structures and must be harmonised before VOSviewer can read them together.
- Generic terms: words such as study, model or approach dominating the map and hiding the substantive topics.
Choosing the right view
Network view
Shows cluster structure. Node size reflects occurrences or citations; link thickness reflects link strength between items.
Overlay view
Colours nodes by average publication year or average citations, which helps separate emerging topics from mature ones.
Density view
A heat-map style display of where research activity is concentrated and where it thins out, useful when arguing for a gap.
The Basic package includes four VOSviewer maps, typically one keyword network in all three views plus one collaboration or citation-based map. The Advanced package includes eight maps across network, overlay and density views for several units of analysis.
Our VOSviewer workflow
Data intake
Send us an existing export, or we design the search and retrieve the records through your institution's licensed access. We check that keyword and reference fields are complete.
Thesaurus building
We review the full term list, merge variants and remove generic or publisher boilerplate terms that carry no thematic meaning.
Network type and threshold
We match the unit of analysis to your research question, test several thresholds and compare how stable the clusters remain.
Layout and clustering
Resolution, minimum cluster size and the attraction and repulsion parameters are adjusted until the map is balanced and clusters are coherent.
Finishing and export
Labels, colours and zoom are refined, then maps are exported as high-resolution PNG plus PDF or SVG, alongside map and network files.
Parameters we report
| Setting | Typical choice | Effect on the map |
|---|---|---|
| Counting method | Full or fractional counting | Fractional counting down-weights documents with many authors or keywords |
| Minimum threshold | For keywords, often 5 to 10 occurrences | Controls node count and legibility |
| Normalisation | Association strength, fractionalization or LinLog/modularity | Changes inter-node distances and cluster shape |
| Clustering resolution | Default 1.00, adjusted when justified | Higher values produce more, smaller clusters |
| Minimum cluster size | Scaled to network size | Prevents trivial one or two node clusters |
| Software version | Current stable release at time of analysis | Version number is recorded for reproducibility |
Reporting VOSviewer settings in your paper
A complete description names the type of analysis, the unit, the counting method, the threshold, how many items met it out of the total, the normalisation method and the clustering settings. An illustrative, hypothetical sentence might read: a co-occurrence network of author keywords was built in VOSviewer using fractional counting, and 142 of 3,870 keywords met the minimum of eight occurrences. Our report supplies every number you need to write this in your own words.
Deciding what each cluster represents, naming it and explaining why clusters connect is the intellectual contribution of the author, and we leave it with you. We also recommend that you place the thesaurus file and parameter log in supplementary material or a public repository, since a growing number of journals ask for exactly that when a paper relies on bibliometric maps.
Integrity commitments
Because you receive the native files, you are never locked in: any later rerun, by you or by a co-author, starts exactly where we left off. Prefer to learn the tool yourself? Start with our VOSviewer tutorial. When you would rather hand it over, see pricing and the order form.