Similar Paper Multi-Platform Search

Problem

Finding similar papers often fails for practical reasons: queries are too narrow, platform coverage is incomplete, results are duplicated, and links cannot be verified. Generating titles, DOIs, or URLs from memory is unreliable; high citation counts do not imply high similarity; paywalls, logins, CAPTCHAs, and rate limits must not be bypassed.

How It Works

The skill first converts the request into a search spec: research question, domain, method, dataset, year range, language, document type, preprint acceptance, and result count. It then builds exact, expanded, and combined query groups, while avoiding unbounded keyword sprawl. By default it targets general academic indexes such as OpenAlex, Semantic Scholar, and Crossref, and adds domain sources like arXiv, PubMed, IEEE Xplore, ACM, or CNKI when relevant. Candidate records preserve title, authors, DOI, PMID, arXiv ID, publisher page, and legal open-access link. Deduplication uses DOI, stable identifiers, and normalized title matching; preprints and published versions are merged while both links are retained. Ranking prioritizes core-concept overlap, abstract coverage, and task/method similarity; citations, recency, and venue quality are secondary. Finally, it verifies authoritative pages and outputs a table plus query, coverage, filters, limitations, and retrieval time.

Boundaries

It is useful for engineering research, literature review, and first-pass screening in algorithm, medical, or engineering topics. It does not guarantee coverage of all databases or global optimality. Failed platforms are marked rather than misrepresented, and when no reliable result exists it says so instead of inventing papers or URLs.