DOME Observatory

The DOME Observatory corpus as of 2026-09-25: 876,324 publications from Europe PMC, each screened for whether it is an AI/ML methods paper (367,348 positive, 502,002 negative, 6,974 undeterminable), with the rationale for each verdict, controlled-vocabulary tags where the enrichment pass has run, citation counts and links to the data each paper cites. Records are LLM-classified by a method validated against a hand-annotated expert benchmark; they are not individually curator-reviewed. Files: the records as JSON Lines, one complete record per line, gzipped -- the same records https://observatory.dome-ml.org/api/export serves; archive-metadata.json, with the counts and the sha256 of every file; and ai-ml-landscape.schema.v1.6.0.json, the record schema (v1.6.0) the records follow. Each version of this record is one corpus release; the concept DOI always resolves to the latest. Licence: the screening decisions, rationales, enrichment tags, vocabularies, schema and statistics this project adds are CC BY 4.0. The bibliographic metadata and abstracts each record carries were not created by this project and are not covered by it; most come from Europe PMC, which should be cited alongside this dataset. Browse and search: https://observatory.dome-ml.org