Build a Literature Review Dataset with OpenAlex

A defensible literature review records how candidates were found, deduplicated, screened, included, and excluded. The OpenAlex Research Scraper can supply a reproducible discovery dataset, but the review protocol still belongs to the research team.

Translate the question into searches

Define concepts, synonyms, publication years, and access criteria before collection. Run concept combinations separately if query provenance matters.

{
  "queries": ["AI safety evaluation", "model interpretability"],
  "fromPublicationYear": 2020,
  "maxResultsPerQuery": 250
}

Archive the exact input, run IDs, collection date, result cap, and raw datasets. Use OpenAlex ID as the primary key and DOI as a secondary cross-database identifier. Create a separate table for query membership if a work appears in several searches.

Add screening fields without changing source data

Keep published title, abstract, authors, institutions, source, topics, dates, and links immutable. Add review columns for title-screen decision, abstract-screen decision, full-text decision, reason for exclusion, reviewer, and timestamp.

Missing abstracts should route to manual lookup rather than automatic exclusion. Open-access status helps find text but is not a quality or eligibility criterion unless the protocol explicitly says so. Citation counts can prioritize reading but can bias against recent or niche work.

Open DOI and publisher pages for included studies and verify metadata. Search additional databases when the protocol requires them. Report source coverage and the number truncated by any cap. A literature dataset is most useful when every inclusion can be traced to a query and every transformation can be reproduced from retained raw rows.