Citation Verifier

Problem

A paper may look properly cited, yet some entries have plausible authors, years, and venues while the title or DOI cannot be found. This is common in generation-assisted writing, where a model may “fill in” references to appear credible, and manual checking is slow. This skill turns that problem into a retrieval task: given a paper or a reference list, determine whether each citation is likely to exist in public academic indexes and report a confidence level.

How It Works

• Extraction: Pull entries from References, Bibliography, or structured citation blocks, recognizing formats such as APA, MLA, Chicago, and BibTeX, then extract title, authors, year, and venue.
• Verification: Query OpenAlex -> CrossRef -> Semantic Scholar in order; if a DOI is present, prefer DOI lookup for exact matching.
• Similarity judgment: Compare the supplied title with titles returned by the APIs using token-set similarity, classifying results as high, medium, low, or not found rather than relying only on whether an API response exists.

Boundaries

This is useful for screening suspicious citations, especially those that may be AI-generated. It depends on the coverage of public APIs, so obscure preprints, internal reports, or very recent publications may not be found. If author or year fields do not fully match, confidence may be moderate. For formal academic-integrity decisions, still verify against original papers, DOI registration records, and publisher metadata.