How to do a bibliometric analysis, step by step

A practical workflow for a bibliometric review: from the research questions and the search string to merged, cleaned data, performance analysis, science maps, and a results section that reviewers accept.

By Bibin XavierUpdated 17 October 202612 min read

What bibliometric analysis is

Bibliometric analysis uses the metadata of publications (authors, sources, keywords, affiliations, references and citations) to describe a research field quantitatively. It complements a systematic literature review: instead of reading every paper in depth, it maps the whole body of work and shows who publishes, where, on what topics, and how the field is connected and changing.

Most studies combine two families of techniques (Donthu et al., 2021):

  • Performance analysis describes productivity and impact: publications per year, top sources, authors, institutions and countries, citation counts and indicators such as the h-index.
  • Science mapping describes relationships: keyword co-occurrence, co-citation, bibliographic coupling and co-authorship networks, thematic maps and their evolution.

Step 1: Define the research questions

Write down what you want to learn before you search. Typical bibliometric research questions are:

  • How has research on the topic grown, and which sources publish it most?
  • Who are the most productive and most influential authors, institutions and countries, and how do they collaborate?
  • What are the main themes, how are they connected, and how have they evolved?
  • What are the intellectual foundations (the most cited references) and the emerging research fronts?

Each question maps onto specific analyses in step 5 and 6, which keeps the results section focused.

Web of Science Core Collection and Scopus are the standard choices because they export complete metadata, including cited references, which co-citation, coupling and historiograph analyses need. PubMed is the reference database for biomedicine but has no citation counts or reference lists. Dimensions, Lens.org and OpenAlex offer broad free coverage.

Build the search string from synonyms of your core concepts, combined with Boolean operators, and restrict it to title, abstract and keywords. Record the database, the exact string, the date of the search and every filter (years, document types, languages). Reviewers will ask for them, and the PRISMA diagram needs the counts.

TITLE-ABS-KEY ( ( "digital health" OR "mhealth" OR telemedicine )
  AND ( "machine learning" OR "artificial intelligence" ) )
  AND PUBYEAR > 2009 AND PUBYEAR < 2026
  AND ( LIMIT-TO ( DOCTYPE , "ar" ) OR LIMIT-TO ( DOCTYPE , "re" ) )

Step 3: Export the records and merge databases

Export the complete record including abstracts, keywords and cited references. The export guide shows the exact clicks for each database. If you searched more than one database, drop all files into Bibliometrics Studio: records are merged and duplicates removed by DOI and by normalised title plus year, keeping the most complete version of each record.

Step 4: Clean and screen the data

  • Check completeness. The Data completeness page shows the share of records missing each field. An analysis built on a field that is mostly missing (for example affiliations in some exports) should be dropped or flagged.
  • Filter. Apply the inclusion criteria: years, document types (often articles and reviews only), languages and topic terms.
  • Merge keyword synonyms. "AI", "artificial intelligence" and "artificial intelligence (AI)" should count as one term. Add synonym groups on the Keyword cleaning page.
  • Remove uninformative terms. Search terms themselves (they appear in every record) and generic index terms such as "human" or "article" add noise to keyword maps.
  • Author name ambiguity. Names are standardised as surname plus initials, so different authors can share a name and one author can appear under two. Check the top authors by hand.

Step 5: Performance analysis

Start with the Main information table (timespan, documents, sources, annual growth rate, average citations, authors, collaboration index, international co-authorship) and the annual scientific production chart. Then report the most relevant sources, authors, affiliations and countries, with their citations and local impact (h-, g- and m-index).

Two classic laws describe concentration. Bradford’s law identifies the core journals that publish about a third of the documents. Lotka’s law shows that most authors publish only once on a topic, while a few are highly productive.

Distinguish global citations (all citations in the database) from local citations (citations from other documents in your collection). A document with many local citations is important to this specific community.

Step 6: Science mapping

  • Keyword co-occurrence network: the topical structure. Clusters are themes; the overlay view colours terms by average publication year to reveal recent topics.
  • Thematic map and thematic evolution: themes classified as motor, niche, emerging or declining, and basic, and how they change across periods.
  • Co-citation analysis: the intellectual base (pairs of references cited together).
  • Bibliographic coupling: current research fronts (documents that cite the same references).
  • Collaboration networks: co-authorship among authors, institutions and countries, and a world collaboration map.
  • Historiograph, reference spectroscopy, trend topics and keyword bursts: the field’s roots and its newest directions.

The science mapping guide explains how each map is built and how to read it.

Step 7: Interpret the results

Numbers alone are not findings. For each cluster or theme, read the most cited documents in it and give it a meaningful name. Link the performance results to the maps: are the most productive countries also central in the collaboration network? Do the emerging themes in the overlay view match the recent bursts? Finish with a research agenda based on niche, emerging and basic themes.

Step 8: Report the analysis

Report the database, search string, search date, filters and the number of records at each step (a PRISMA flow diagram), the software and its version or access date, how duplicates and synonyms were handled, the parameters of each map (items, thresholds, normalisation, clustering), and the main tables and figures. Bibliometrics Studio generates the PRISMA diagram, a Word report and a methods paragraph that covers these points.

The BIBLIO checklist (Montazeri et al., 2023) lists the minimum items for bibliometric reviews in the biomedical literature and is useful in any field. Please cite the software you used. The citation for Bibliometrics Studio is on its home page and in the app.

References

  • Aria, M., & Cuccurullo, C. (2017). bibliometrix: An R-tool for comprehensive science mapping analysis. Journal of Informetrics, 11(4), 959–975.
  • Cobo, M. J., López-Herrera, A. G., Herrera-Viedma, E., & Herrera, F. (2011). Science mapping software tools: Review, analysis, and cooperative study among tools. Journal of the American Society for Information Science and Technology, 62(7), 1382–1402.
  • Donthu, N., Kumar, S., Mukherjee, D., Pandey, N., & Lim, W. M. (2021). How to conduct a bibliometric analysis: An overview and guidelines. Journal of Business Research, 133, 285–296.
  • Montazeri, A., Mohammadi, S., Hesari, P. M., et al. (2023). Preliminary guideline for reporting bibliometric reviews of the biomedical literature (BIBLIO): A minimum requirement. Systematic Reviews, 12, 239.
  • Page, M. J., McKenzie, J. E., Bossuyt, P. M., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ, 372, n71.
  • Zupic, I., & Čater, T. (2015). Bibliometric methods in management and organization. Organizational Research Methods, 18(3), 429–472.

Related guides

Export from Scopus, WoS and PubMedThe exact export settings for each database Science mapping guideCo-occurrence, co-citation, coupling and thematic maps Meta-analysis calculatorPool effect sizes from the studies you review