Bibliometrics quantitatively maps a field's structure and evolution. In AEC it is widely used to frame review papers. Used well it is rigorous; used badly it is "descriptive statistics dressed as insight" — push for interpretation, not just colourful network pictures.
1. Two complementary analysis families
Performance analysis (productivity & impact): publications per year, most-productive/most-cited authors, institutions, countries, journals; citation counts; h/g-index; Bradford's law (core journals); Lotka's law (author productivity). Science mapping (structure & dynamics):
- Co-citation (of references/authors/journals) → intellectual base / foundational works.
- Bibliographic coupling (shared references) → current research fronts / active clusters.
- Co-word / keyword co-occurrence → conceptual/thematic structure & trends.
- Co-authorship → social/collaboration structure (authors, institutions, countries).
- Thematic evolution / overlay timelines → how themes rise, merge, fade (CiteSpace bursts; Bibliometrix thematic map quadrants: motor / niche / emerging-declining / basic themes).
2. Workflow
- Define scope & query (same rigor as a systematic search — see
systematic-review). Bibliometrics is only as good as the export. Document the database, exact query, filters, and export date. - Choose ONE primary database for the structured analysis (Scopus or Web of Science) because citation/ reference metadata is not mergeable cleanly across them; you may report coverage from both. OpenAlex/ Dimensions are open alternatives. Note coverage limits (conferences, books, non-English under-indexed).
- Export full records + cited references (Scopus: CSV/RIS/BibTeX with references; WoS: tab-delimited/ plain text "Full Record and Cited References").
- Clean the data — this is where credibility is won or lost:
- Disambiguate author names (initials collisions), unify institution/country variants.
- Merge keyword synonyms & spelling variants (BIM / "building information modeling" / "building information modelling"; "life cycle assessment" / LCA) via a thesaurus file (VOSviewer supports one).
- Remove stop-words / overly generic terms.
- Analyze with a tool (below). Choose the analysis to the question — don't run all of them by reflex.
- Interpret — name and explain clusters, link to the narrative, identify gaps and a research agenda.
Validate clusters against domain knowledge (
aec-domains); a cluster the field wouldn't recognize is a data-cleaning artifact.
3. Tools
| Tool | Best for | Notes |
|---|---|---|
| VOSviewer | Co-occurrence, co-citation, coupling, co-authorship maps; clean visuals | Free; uses association strength; supports thesaurus; great for figures |
| CiteSpace | Temporal dynamics, burst detection, structural variation, pivotal nodes | Free (Java); steep learning curve; strong for evolution & emerging trends |
| Bibliometrix / biblioshiny (R) | Full pipeline incl. performance analysis, three-fields plot, thematic map/evolution, Lotka/Bradford | Open, reproducible, scriptable; biblioshiny = GUI |
| OpenAlex / pyalex / litstudy / metaknowledge | Programmatic, open, no subscription | Reproducible pipelines; good for transparency |
| Litmaps / Connected Papers / ResearchRabbit | Visual citation discovery & snowballing | Discovery aids, not formal analysis |
| Gephi | Custom network layout/metrics | When you need bespoke graph analytics |
4. Reporting standards
- Follow guidance from Donthu et al. (2021) and Aria & Cuccurullo (Bibliometrix). For the review wrapper,
combine with PRISMA flow (see
systematic-review). - Report: database, query string, filters, time span, export date, record count, software + version, cleaning steps (thesaurus decisions), parameters (min. occurrence thresholds, clustering resolution), and units (full vs fractional counting).
- State thresholds explicitly (e.g., "keywords with ≥5 occurrences"; "authors with ≥3 documents") — these drive every figure and must be reproducible.
5. Metrics & laws cheat-sheet
- Bradford's law — few core journals hold most papers (justifies core-source selection).
- Lotka's law — author productivity is highly skewed (few prolific authors).
- Price's law — ~half of papers come from √N of authors.
- h-index / g-index — productivity+impact; age- and field-biased.
- Total/average citations per item, citing articles, FWCI — impact; field-normalize when comparing.
- Total Link Strength (VOSviewer) — node connectedness in the map.
- Betweenness centrality / burstness (CiteSpace) — pivotal & surging nodes.
6. Interpreting the science map (turn pictures into findings)
- Co-citation clusters = the field's intellectual foundations → write the "theoretical roots."
- Coupling/keyword clusters = current fronts → write "active research streams."
- Sparse links between clusters = fragmentation / integration opportunity → a gap.
- Recent keyword bursts / emerging-theme quadrant = trends to position new work against.
- Peripheral isolated nodes = niche or nascent topics → potential novelty space.
- Country/institution networks = collaboration gaps, under-represented regions (a legitimate AEC gap).
7. Pitfalls to flag
- Merging Scopus + WoS citation data naively (double counting, mismatched references).
- No data cleaning → inflated/duplicated authors & keyword noise → meaningless clusters.
- Treating cluster colors as conclusions without interpretation.
- Equating citation count with quality (DORA caution).
- Ignoring database coverage bias (English, journal-over-conference, recent-over-old).
- Over-reading small networks; report sensitivity to thresholds.