How to Do Bibliometric Analysis: A Step-by-Step Guide for PhD Researchers

To do bibliometric analysis, define a focused question, collect suitable bibliographic records, screen and clean the dataset, select appropriate techniques, and interpret the results through close reading. This guide explains how to document each stage for a PhD thesis or research manuscript.

What Is Bibliometric Analysis?

Bibliometric analysis examines publication metadata and relationships among documents, authors, sources or terms. It can show how a selected literature collection has developed and how its contributors and ideas connect. It does not, on its own, establish whether the evidence in a paper is reliable.

Separate two aims: describing the collection, such as publications by year, and mapping relationships, such as collaboration or shared references. Start with the aim rather than choosing attractive charts first.

1. Define Your Research Question and Scope

Replace a broad topic such as “artificial intelligence” with a question you can investigate. For example: “How have research topics and collaboration patterns in AI-assisted higher education changed between 2015 and 2025?” This is an illustrative research plan, not a completed study.

Write down the subject boundaries, publication years, languages, document types and units of analysis. Explain why each restriction fits the question. A manageable dataset is helpful, but arbitrary exclusions can distort the field you are describing.

2. Choose a Database and Build Your Search

Choose a source with suitable disciplinary coverage and the metadata your analysis requires. Check export permissions and whether cited references are available. Bibliometrix’s official data-import guide describes a search, refinement and export workflow; always match the export to your installed software version.

For the illustrative topic, a starting Boolean expression might be:

("artificial intelligence" OR "machine learning") AND ("higher education" OR universit*)

This expression is not validated. Adapt field tags, phrase handling and truncation to the database. Inspect results for irrelevant meanings and missed terminology. Check whether known relevant papers are retrieved, then refine and record the final query.

Save the database name, exact search expression, search date, filters and initial result count. If you search multiple databases, retain each source separately before merging.

3. Export and Screen the Records

Export a compatible file containing the fields required for the planned analyses. These may include title, authors, year, source, DOI, affiliations, keywords, abstract and cited references. A reference-manager export can omit fields needed for science mapping, so check a small import before exporting a large collection.

Apply the documented eligibility criteria and record exclusion reasons. Keep the untouched original export and create a separate working copy. Record counts at collection, deduplication, screening and final inclusion should reconcile. Do not describe bibliometric screening as a full systematic review unless the study actually follows that methodology.

4. Clean the Dataset Without Erasing Meaning

Check exact and probable duplicates using identifiers and bibliographic details. Review spelling variants in keywords and institutional names. Avoid merging authors simply because their initials match, or collapsing distinct concepts into a single term.

For example, “AI” may refer to artificial intelligence in one context and a different concept in another. A replacement dictionary needs human review. Record each merge, exclusion and unresolved ambiguity in a cleaning log. Use our bibliometric dataset cleaning checklist to organise this stage.

5. Match the Analysis to the Question

Pick a small set of analyses that address the question. Explain your choices and recognise what the dataset cannot answer. A citation count reflects recorded attention within a source and time window; treat it cautiously when comparing older and newer publications.

6. Run the Analysis and Keep the Settings

VOSviewer constructs and visualises bibliometric networks. Bibliometrix offers an R-based workflow for bibliographic analysis. Biblioshiny supplies a graphical interface for users who prefer to work without writing R code. See our VOSviewer vs Bibliometrix comparison before deciding.

Record software versions, analysis units, counting method, thresholds, normalisation and clustering settings. Explain why a threshold is appropriate and check whether a nearby setting changes your main interpretation. Keep the retained-item count alongside each network figure.

7. Read the Maps Alongside the Papers

Inspect the labels, associated records and representative papers in each cluster. Name a cluster only after understanding its subject matter. Two connected keywords show a relationship in the selected records; they do not demonstrate causation or establish that one topic is more valuable.

An isolated node or small cluster is not sufficient proof of a research gap. Verify whether the pattern results from query design, terminology, missing metadata or actual limitations in the literature. Then use close reading to develop a defensible research argument.

8. Write a Transparent Methods and Results Section

Report what you searched, how you selected and cleaned records, what you analysed and how you interpreted the outputs. Keep descriptive results separate from explanations and proposed research directions.

• Methods: source, search date, query, criteria, cleaning and software settings.
• Results: relevant tables and maps, with clearly defined measures.
• Discussion: interpretation supported by reading and context.
• Limitations: source coverage, language restrictions, citation timing and metadata quality.
• Reproducibility: share scripts, cleaning rules and permitted files, subject to licensing.

Common Mistakes to Avoid

• Choosing the dataset after seeing a preferred result.
• Treating every cluster as a separate research gap.
• Combining database citation counts without explaining their differences.
• Showing maps without methods, labels or interpretation.
• Reporting software defaults as if they were justified research decisions.

Frequently Asked Questions

How many papers do I need?

There is no universal minimum that makes every study valid. Consider scope, metadata quality and whether the collection supports the intended analyses. A small specialised collection may require a different approach from a broad field.

Does bibliometric analysis replace a literature review?

It can help organise and explore a literature collection, but critical synthesis still requires reading and assessing relevant studies.

Can I do bibliometric analysis without coding?

Yes, graphical tools can support many tasks. You still need to make and document methodological decisions rather than treating software output as an explanation.

Need Help Planning Your Study?

Our bibliometric analysis services can be discussed around your research question, dataset and required outputs. Email info@phdscholarsdesk.com with your topic, available export and deadline to request a scoped quote.