Searching the databases from the app
Write the search string once; the app writes it for each database, counts the hits, retrieves the records and logs every search for the PRISMA-S appendix. Which databases it searches, which you search by hand, keys, and citation searching.
On this page
The Import step can run your searches for you. Every search, whether the app ran it or you imported a file, becomes one entry of the search log, which the paper reports as its PRISMA-S appendix.
1. Write the search string once
In Config or on the Import step, write the string the usual Boolean way:
("tiny machine learning" OR TinyML) AND (quantiz* OR pruning)- Words and "quoted phrases"; a trailing
*for truncation. AND,OR,NOT(orAND NOT) and parentheses.- Optional fields:
title:,abstract:,keywords:,tiab:(title and abstract). Without a field, a term is searched in the title, abstract and keywords wherever a database allows. - Strings copied from Scopus, Web of Science, PubMed or IEEE Xplore are read too (for example
TITLE-ABS-KEY( ),TS=( ),"x"[tiab]).
2. Check each database's query
The app writes the string in each database's own syntax and says what a database cannot express (for example NOT, truncation or field limits). When the rules cannot serve a database fully, the AI writes that query. You can edit every query before anything runs.
3. Count, then retrieve
- Each database is counted first, so you see the hits before retrieving.
- A background job retrieves up to a cap per database: 2,000 by default, 10,000 at most, with a warning when a database finds more.
- Each database's results are saved as one search in the search log, exactly like an imported file: the source, the interface (API), the date and time, the exact query and request, the limits, the count the database reported and the records retrieved.
- Duplicates are found as for any import.
Run again repeats a search with the same query and limits and adds only what is new since the last run.
Which databases
| Searched by the app | What it needs |
|---|---|
| OpenAlex (Boolean in titles, abstracts and keywords) | Nothing; a free key raises the daily budget |
| Semantic Scholar (bulk Boolean search) | Nothing; a free key gives the site its own rate |
| arXiv (title and abstract fields) | Nothing |
| DBLP (titles only; abstracts added from OpenAlex or Semantic Scholar by DOI) | Nothing |
| Crossref (relevance search, not Boolean: the paper says so) | Nothing |
| Europe PMC and PubMed (for AI in health) | Nothing; an NCBI key raises PubMed's rate |
| CORE | Nothing; a free key raises the limits |
| Springer Nature | A free key |
| IEEE Xplore (metadata and abstracts) | Your own free key from IEEE's developer site, with IEEE's terms shown and accepted |
| Scopus | An Elsevier key used from inside a subscribing institution's network (or VPN); otherwise searched by hand |
Searched by hand
Google Scholar, the ACM Digital Library, Web of Science, ScienceDirect, and IEEE Xplore without a key do not allow searching by software, and the app never scrapes them. For each, the app writes the query in its syntax, opens the search there with the years set, lists the export steps, and fills in the import form for the file you export:
| Database | How to export |
|---|---|
| IEEE Xplore | Select all on the page, then Export, Citations, format RIS, "Citation and Abstract". Up to 1,000 results at a time: export in batches (for example by year) |
| Scopus | Through your institution: select all, then Export, RIS (or CSV) with citation information and the abstract; up to 20,000 at once |
| Web of Science | Through your institution: Export, "RIS (other reference software)" or plain text, records 1 to 1,000, Full record; then the next 1,000 |
| ACM Digital Library | Select all, then Export citations, BibTeX or EndNote; show 50 per page and export each page |
| ScienceDirect | Through your institution: select all, then Export citation to RIS; up to 100 at a time |
| Google Scholar | At most 1,000 results and no bulk export: save results to "My library" and export it, or use a reference manager. Report it as a supplementary search |
Then import each file on the Import step: the source, query and limits are already filled in.
Keys
Add your keys (for example IEEE Xplore's) in the app. Keys are stored encrypted and shown only masked: each person's in their own account, the site's in System. Requests to databases identify the site, never your personal e-mail address.
Citation searching (snowballing)
From the included studies, the app follows:
- their references (backward), and
- the works that cite them (forward),
through OpenAlex, and Semantic Scholar for studies OpenAlex does not know. What the review does not have yet is imported as one search per direction under "other methods" and screened by title and abstract like the database records. The flow diagram gives other methods their own boxes, and the counts still reconcile.
In the paper
The methods name every source, the interface, the dates, the cap where a search hit it, Crossref's relevance search and DBLP's titles, and the queries the AI wrote. The search appendix gives each search's query, request, limits and counts.