Frequently Asked Questions

The platform offers three search modes:

1. Experts by topic

   Enter terms related to your topic of interest. The platform searches publication titles and lists the researchers associated with the matching works.

2. Publications by topic

   Enter terms to find works by their titles. Results show publication records and associated researchers, with links to their profiles.

3. Experts by name

   Enter a person’s name or part of their name. This search queries researchers’ names, regardless of the topics of their publications.

In topic searches, terms are queried in publication titles, and the year range, when specified, restricts the works considered. The search does not query the full text of the works.

In Experts by topic, frequency indicates the number of matching publications for each researcher; the initial sorting prioritizes higher frequencies. In Publications by topic, works are initially listed from the most recent year to the oldest. In Experts by name, the initial sorting is based on the researcher’s total number of publications.

List filters refine results using the fields available in each mode, such as subject area, affiliation, location, publication type and ASJC category. Sorting and page-size controls allow you to reorganize the results. In the publication list, the Record filter also queries authors, venues and identifiers.

To learn more about how the Phase I search terms were developed, consult the document shared below. It describes the development and organization of the interdisciplinary vocabulary used to identify researchers, explains how to interpret the searches, and provides the complete list of 327 expressions.

A text match is evidence for examining a CV, not an automatic definition of a person's specialty. Selection applies the Phase I criteria; in Phase II, publications from the selected CVs feed the platform's public search.

Description and list of search terms — Phase I (Word, in Portuguese)

The most recent collection of Lattes CVs took place in September 2026.

The abbreviations below describe the output recorded for each researcher. Freq refers to works matching the search, while the other counts summarize the recorded output.

• Freq. Number of matching publications for the researcher based on the terms searched in titles. Available in expert searches by topic; it takes into account the period specified in the query.
• Eve. Number of fulltext articles published at events.
• Jou. Number of fulltext articles published in journals.
• Cha. Number of published book chapters.
• Boo. Number of published books.
• N. Total number of publications (Eve + Per + Cap + Liv).

Each result displays a bibliographic record and the specialists in the database associated with the work. The same fields also appear in profile publication lists.

• Record. Bibliographic information about the work: title, authors, publication venue, year and other available data, such as volume, issue, pages and identifiers.
• Specialists. People in the database associated with the publication, with links to their profiles. This is not necessarily the complete list of the work’s authors.
• Publication venue. Journal, event or book in which the work was published.

Identifiers help locate the publication or the venue where it was published. Links let you look up the work, while exporting lets you save the results in a spreadsheet.

• DOI (Digital Object Identifier). A persistent identifier for a publication. When available, the DOI link leads to the work’s page.
• ISBN (International Standard Book Number). An identifier for a book edition.
• ISSN (International Standard Serial Number). An identifier for a serial publication, such as a journal; it does not identify an individual article.
• Google Scholar. A link to search for the work’s title in Google Scholar, shown when the record has no DOI.
• CSV (Comma-Separated Values). A tabular file format used to export results, which can be opened in spreadsheet programs.

ASJC stands for All Science Journal Classification, the subject classification system used by Scopus. The platform adopts a catalog of 334 categories and assigns one category to each publication whenever possible. This classifies a work by its bibliographic content; it is neither a quality assessment nor a statement of its authors' expertise.

The reference catalog is available in the Scopus documentation: All Science Journal Classification Codes (ASJC).

On the platform, categories are organized into five subject groups: HS (Health Sciences), LS (Life Sciences), MU (Multidisciplinary), PS (Physical Sciences) and SH (Social Sciences & Humanities). A publication's subject area is determined by the group of its ASJC category. Select a group below to explore its categories.

The predominant ASJC category is the most frequent among classified publications. The percentage uses classified publications only; the displayed subject area belongs to that category.

In specialist searches and member and friend directories, ASJC filters select the displayed predominant category and that category's subject area. The calculation considers all of the person's classified works, not only those matching the search terms. In a tie, the first category in alphabetical order is displayed.

Classification was developed in stages, using titles and, when available, publication venues. In earlier versions, about 40,000 records were labeled automatically with gpt-3.5-turbo-0613 through the OpenAI API (Application Programming Interface). The resulting title, venue and ASJC category triples were used to fine-tune a multilingual BERT (Bidirectional Encoder Representations from Transformers) model, applied to about 2.2 million publications.

In November 2024, the Llama 3.1 model, with 70 billion parameters, was used to classify about 600,000 publications added to the database. These figures describe historical stages and should not be added together to estimate the current database size.

In the October 2026 update, 1,581,617 historical classifications were reused through exact bibliographic metadata matching. For the remaining records with a title, the Gemma 4 12B model (gemma4:12b), with approximately 12 billion parameters, was run locally through Ollama. Inference was performed in batches, with a hierarchical subject-area and category selection step for cases requiring another attempt, resulting in 1,343,865 additional classifications.

The October 2026 database contains 2,925,940 publication records: 2,925,439 received an ASJC category and 501 remained unclassified because of a missing title or abstention by the procedure. The subject area is obtained from the assigned category's catalog entry; it is not inferred separately. Results are automatic and may contain errors: they do not constitute individual human validation or guarantee scientific accuracy. Unclassified publications are excluded from the calculation of the predominant category and its percentage.

• Devlin, J. et al. BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018.
• Official documentation of the OpenAI ChatGPT API.
• Multilingual BERT model without case sensitivity differentiation.
• Llama model 3.1 with 70 billion parameters.
• Gemma 4 12B model used in the 2026 update (Ollama).

R. Lent, D. Foguel, M. Z. Guimarães, D. Botaro, J. P. Mena-Chalco. Science for Education network: The Brazilian proposal. In book: Developing Minds in the Digital Age: Towards a Science of Learning for 21st Century Education. OECD Publishing, Paris. 2019.

Contact us via e-mail contato@cienciaparaeducacao.org.