Bibliometrics and research output analysis

With a bibliometric analysis that quantifies the state of the field: how much is published, on what topics, who leads and which niches are under-explored. It applies when an evaluator for a competitive funding call, such as the Carlos III Health Institute (ISCIII), Horizon Europe or the ERC, asks for evidence of relevance and novelty, and a sentence about "growing interest" does not provide it. The analysis maps output by year, author, institution and country, builds collaboration networks and generates thematic maps that classify each area as established, emerging or declining, so that the claim "this is an emerging field" moves from rhetoric to data. UpSet plots of MeSH terms add the detection of missing thematic combinations, which can point to unexplored lines of research. What you receive is quantitative evidence that helps support the case for novelty, not a promise that the proposal will be funded, which depends on many other factors.

The analysis draws on the major bibliographic databases, such as PubMed, Web of Science and Scopus, and the search string is defined by the service together with the client, rather than handed over ready-made by the client. This applies to any bibliometric project, where the quality of the map depends directly on how the corpus is delimited. A bibliometric search aims for broad, coherent coverage of the field, with different criteria from those of a systematic review, which pursues exhaustiveness under strict criteria. The service translates the client's question into a search string, downloads the records, removes duplicates and standardises author and institution names before the analysis. The client contributes knowledge of the field and validates the scope; the service builds and runs the search strategy and prepares the corpus. Defining the search string well is what separates a useful, defensible map from one skewed by a poorly framed search.

A centrality-density thematic map classifies the themes of a field into four quadrants according to their structural relevance (centrality) and their degree of internal development (density). It applies when you want to understand not only how much is published, but how the themes are organised and where the field is heading. Themes are classified as motor themes, basic and transversal themes, highly specialised themes or, where the temporal reading allows it, emerging or declining themes. The map is built from keyword co-occurrence and thematic clustering techniques, depending on the approach chosen, to identify the thematic groups. One caveat: a theme's position depends on the corpus and the period analysed, so it is read as a snapshot of a delimited field, not as absolute truth. What it provides is an actionable reading of which themes to prioritise, consolidate or explore.

They add a reading of the content and tone of the corpus that bibliometric counts do not capture: what exactly is being discussed and with what orientation, beyond how much is published. This applies when you are interested in the thematic and emotional content of the literature, not only in its output and collaboration metrics. LDA topic modelling proposes latent thematic groupings in the abstracts without supervision, which do not always match the keywords the authors declared; sentiment analysis with lexicons such as NRC or Bing approximates the patterns of polarity or emotion in the language of the abstracts. One limit applies: these techniques describe lexical patterns in the corpus, not judgements on the quality or truthfulness of the research, and latent themes need expert interpretation to be useful. What they add is a content layer (what is said and how) that complements the structure classic bibliometrics provides.

Bibliometrics is used to map the structure, evolution and dynamics of a research field: who publishes, on what, with whom and how it changes over time. Meta-analysis fits when the question is quantitative and about a specific effect: combining the numerical results of several studies to estimate a pooled effect size. They do not answer the same question: bibliometrics analyses the literature as an object, that is, its output and its structure; meta-analysis synthesises the results of the literature to answer a clinical or epidemiological question. The search strategy also differs: in bibliometrics, the search string is built for broad coverage of the field; in meta-analysis, study selection follows explicit eligibility criteria and aims for exhaustiveness. They can be run in sequence, for example with a bibliometric map before deciding the scope of a synthesis, but they are different services.

It does not replace it: bibliometrics is often the preliminary step that prepares and sizes a systematic review, not a substitute for it. This applies when a team is about to undertake a review and first needs to understand the terrain: exploring terms, authors, journals and the volume of literature before settling on a review strategy. The bibliometric map offers that quantitative overview of the field without reading every article, which helps narrow the scope before investing the effort of a full review. The limit is clear: bibliometrics describes the structure of the literature; it does not critically appraise the evidence or draw conclusions on the efficacy of an intervention, which is the job of the systematic review and, where applicable, the meta-analysis. One maps the field; the other appraises the evidence it contains.

Only scientific literature: the NLP in this service is applied to abstracts and bibliographic records, not to health records or patients' clinical text. This limit needs stating because "text analysis in health" can be read in two very different ways. Here the corpus consists of a field's abstracts, to which topic modelling and sentiment analysis are applied to characterise the themes and tone of the research output. Natural language processing of health records, with entity extraction, SNOMED-CT coding and clinical text mining, is a separate line of work, with its own sources, privacy framework and validation, and is not part of this service. The distinction is not a nuance: the sources, the risks and the methods all change. If you need to analyse clinical text, you will find a clear boundary here, not a vague promise.

A quantitative map of the field, documented and reproducible: descriptive and temporal analysis of the output, co-authorship and institutional networks with centrality metrics, centrality-density thematic maps, MeSH term analysis and, when the project includes it, NLP on the abstracts (sentiment and topic modelling). This applies if you need a snapshot of the state of a field to justify a line of research, prepare a review or assess a group's position. The visualisations come with their interpretation, not as loose outputs: networks, thematic maps, UpSet plots, term visualisations, thematic evolution and lexical patterns. The analysis is documented in Quarto with reproducible code, so it can be reviewed and updated. The client contributes knowledge of the field and the question; the service builds the corpus, analyses it and interprets it. What you receive is an actionable map, not a dump of bibliographic metrics.