Specify the question and unit of analysis
Decide whether you are examining documents, authors, institutions, countries or terms. Describe the period and disciplinary boundary. A publication-trend question and a collaboration question need different analyses; do not select every available visualization merely because the software offers it.
Keep a reproducible search and export record
Record the database, search string, fields searched, date, filters and export format. Save the original permitted export and document how records were combined. Coverage, language and document-type choices shape what the analysis can show. Check the database licence before sharing a dataset.
Clean identities and terms with a decision log
Review duplicate records, name variants, institutional changes and ambiguous keywords. Preserve a mapping from original to cleaned labels. Explain how missing identifiers and references were treated. Do not merge two authors simply because their names match.
Choose and report the network settings
Tools such as VOSviewer can help construct and inspect bibliometric networks. State the relationship being mapped, inclusion thresholds, counting approach and settings used. Keep the exported network, cleaning record and software version so the result can be checked.
Interpret patterns without overstating them
Read representative papers when naming a cluster and explain alternative interpretations. Citation frequency does not by itself establish research quality or social benefit. Discuss database coverage, publication age and sensitivity to your cleaning or threshold decisions. A bibliometric overview can complement a critical review, but it does not replace reading the evidence.