Bibliometric analysis with VOSviewer and Biblioshiny

What is bibliometric analysis, and how is it different from a systematic review?

Here every paper becomes one row in a spreadsheet. Performance analysis is the counting part, like papers per year, the authors and journals who publish most, and the most cited papers. Science mapping then checks the links. Who writes with whom, which references and keywords keep turning up together, and how these clusters shift over time.

A systematic review starts from one narrow question and screens studies against criteria, then extracts, appraises and combines what they found. Our systematic review guide has that method. In plain words, a systematic review gives you what the evidence says on a question. A bibliometric analysis gives you how the field is organised, and it cannot say whether any one finding is correct.

For this reason the stronger bibliometric papers also read the content. Suppose your maps show four keyword clusters. Read the central papers of each and explain what that strand is arguing. Our guide on writing a review article compares this approach with the other review types. Bibliometrics is meant for topics where there are hundreds or thousands of papers. If your topic has only sixty, read all of them and write a structured review.

How do you write and record a search string for Scopus or Web of Science?

If the reviewer has doubt about your search, he will doubt every map also. So build it with care.

1. Split the topic into concepts. In fintech for small firms, “fintech” is one concept and “small firms” is the second.
2. Collect synonyms for each concept, with abbreviations and both spellings, and use the wildcard * so other endings also get picked. Synonyms go with OR, and between concepts you put AND.
3. Choose the fields. In Scopus, TITLE-ABS-KEY searches titles, abstracts and keywords, and DOCTYPE, LANGUAGE and PUBYEAR restrict type, language and years. In Web of Science, the TS= tag is for searching the topic fields.
4. Check the search with papers you already know. List ten central papers. If the search misses one, find the term that paper uses and add it.
5. Note down the exact string, database, date and number of results on the same day. New records come in daily, and without the date you also can’t rerun it.

For the fintech example, the Scopus string can look like this:

TITLE-ABS-KEY ( ( fintech OR "financial technolog*" OR "digital payment*" ) AND ( sme* OR msme* OR "small business*" OR "small firm*" ) ) AND DOCTYPE ( ar OR re ) AND LANGUAGE ( english )

For Web of Science, put the same terms under TS= and select document type and language from the refine panel. The Scopus Support Center’s advanced search page lists field codes, proximity operators like W/3, and the order in which operators apply. To decide between the two databases, see our comparison of Scopus and Web of Science.

How do you export, merge and clean Scopus and Web of Science data?

Tick all fields while exporting, cited references also, or citation, coupling and co-citation analysis will not run.

These limits keep changing. The VOSviewer manual of October 2023 gives 500 records for Web of Science and 2,000 for Scopus, lower than the vendors’ 2026 help pages. So check the vendor page on the day of export, not some tutorial. When your results cross the limit, export them in batches (1–1,000, 1,001–2,000 and so on) and put the date in each file name.

Merging and removing duplicates

VOSviewer reads many files from one database and removes duplicates on its own, but it cannot combine different databases. For merging Scopus with Web of Science, use bibliometrix in R. First import each set with convert2df() (for example dbsource = "scopus", format = "csv" and dbsource = "wos", format = "plaintext"). Then pass both data frames to mergeDbSources(). Duplicates get thrown out by it while joining the two sets. Write down how many records came from each database and how many duplicates got removed.

Scopus and Web of Science format cited references differently. So before co-citation on the merged file, search a few famous references and see that each comes once only, not twice with slightly different initials. If they are coming twice, do the reference-based analyses on one database only and explain the reason in your paper.

Screening and cleaning

Read the titles, and abstracts where the title is unclear, and remove off-topic records. Then clean the names. “Sharma, R.” and “Sharma, Rakesh” can be the same author. “SME”, “SMEs” and “small and medium enterprises” can be the same idea. In VOSviewer this is handled with a thesaurus file. The file is plain text, tab-separated, with two columns headed label and replace by. On each line you put the label from the map and the label it should become. For a word like “study” leave “replace by” blank and VOSviewer drops it. Keep the file for your supplementary material.

How do you run co-authorship, co-citation, coupling and keyword maps in VOSviewer?

Click Create, then “Create a map based on bibliographic data”, then “Read data from bibliographic database files”, and select your files. The wizard then asks for the type of analysis and the unit.

Many scholars confuse coupling with co-citation. Bibliographic coupling connects two papers when both of them cite the same work. So it groups recent papers by the base they share. Co-citation connects two works when later papers cite both of them together, and that way it shows which classics the field is standing on.

Counting method and thresholds

By default VOSviewer uses full counting. If you choose fractional counting, a paper having many authors gives less weight to each link, and a few big-team papers will not take over the co-authorship map. Mention in the method which one you have used.

The next screen wants a minimum, like keyword occurrences or documents per author. This threshold decides what comes on the map, so it is part of your method and needs reporting. Say how many items crossed it, and check whether the main clusters hold at one lower and one higher value. You can write it like this: “clusters were stable at thresholds of 3, 5 and 10; we report 5, which retained 84 of 2,316 keywords.”

Reading the maps

In the map, circle size shows the weight of an item and colour shows its cluster. Nearby items are mostly related, but take the distance as rough only. In the overlay visualisation the colours follow a score like average publication year, so newer topics stand out. VOSviewer only numbers the clusters. Naming each one from its items is your work.

How do you use Biblioshiny for Lotka’s law, Bradford’s law and thematic maps?

Install R and run install.packages("bibliometrix") one time. After that, in every session, run library(bibliometrix) and then biblioshiny(). A browser tab opens with the app. Load your export or merged file and apply the filters you reported for the search.

Lotka’s and Bradford’s laws

Lotka’s law deals with how many papers each author writes. As per Lotka’s 1926 form, the number of authors with n papers is roughly 1/n² of the number of authors with one paper. That means most authors write just a single paper. Bradford’s law is about how your papers are scattered across journals. The sources get divided into zones, each zone holding about one third of the papers, and the first or core zone has only a few journals in it. Biblioshiny plots both laws for you. Present them as a description of your dataset. For reviewers a table of top authors is only background informations, and they will not count it as your contribution.

Thematic map and three-field plot

The thematic map places the keyword clusters on two axes. Centrality on one axis tells how strongly a theme links to other themes. Density on the other shows how developed it is inside. The four quadrants are read like this. Motor themes are high on both. Niche themes are dense but on the side. Emerging or declining themes are low on both, and basic themes are central but less developed. Whether a theme in the lower-left corner is emerging or dying out, only reading the recent papers can tell you. The three-field plot connects three fields, for example authors, keywords and sources. Ten or fifteen items per column is enough, beyond that the lines get tangled.

You must cite the software also. For bibliometrix it is Aria and Cuccurullo (2017), Journal of Informetrics 11(4), 959–975, and for VOSviewer it is van Eck and Waltman (2010), Scientometrics 84(2).

Most scholars make their network figures in VOSviewer and take the statistics and thematic maps from Biblioshiny. CiteSpace takes more effort to learn. It is worth that effort when your research question is about how a field has changed over time.

How should you report a bibliometric analysis in a paper or thesis?

Write your method section in such a way that another researcher can run it again and land on the same dataset.

A PRISMA-style flow diagram. PRISMA 2020 was made for systematic reviews. Even so, its flow diagram is the clearest way to show what happened to your records. Take the “databases and registers” template and adapt it. Show the records from each database, the duplicates removed, the records screened, the records excluded with the reasons, and the final dataset. Mention in the text that it is adapted.

A table of search parameters, similar to this one:

Put your research questions before the results section. Then arrange the results question by question, and not in the order of software menu.

Why do reviewers reject bibliometric analysis papers?

Editors are getting bibliometric papers all the time, and one that reads like pasted software output is easy to send back. These objections keep coming up:

• Maps with no interpretation. Every figure needs a paragraph on what the clusters mean and which papers are at their centre.
• Thresholds with no rationale. Reviewers see “minimum 5 occurrences” and ask why 5, and whether clusters look the same at other values also.
• One database, no justification. If you used only Scopus, say why, and what you may have missed.
• Descriptive only. Papers per year, top countries and top journals, with no research questions and no conclusion about the field.
• No screening. Off-topic papers always creep into a keyword search and later sit on the map as strange clusters.

Most of these problems have the same fix. Frame two to four research questions that your maps are able to answer, and discuss about every figure in the light of those questions. Read the central papers of each cluster with a critical eye, and our guide to critiquing an article can help with that. End with a research agenda that comes out of what you found. If you can’t tell your guide in one line what a map shows, drop that map.