Reference Authenticity Verification

Problem

Reviewing references in papers, proposals, or technical documents usually requires checking titles, author order, venues, years, pages, and DOIs one by one. Manual checks are slow, and they can miss subtle issues such as altered titles, swapped authors, broken DOIs, retracted articles, or predatory journals. This skill turns a reference list into structured, verifiable entries and produces an auditable review result.

How It Works

The workflow covers extraction, search, verification, anomaly classification, and report generation. It parses mixed Chinese-English citation styles such as APA, MLA, Chicago, and GB/T 7714, then extracts title, authors, venue, year, volume, issue, pages, and DOI/ISBN. Search starts with general sources like Google Scholar, Semantic Scholar, CrossRef, and PubMed, then falls back to discipline-specific databases such as IEEE Xplore, ACM, arXiv, CNKI, and Wanfang. Entries with a DOI are checked against the DOI resolution result. Verification covers existence, title consistency, author accuracy, venue/year/page accuracy, and publisher or conference legitimacy. Problems are labeled with codes such as E-FAKE, E-AUTHOR, E-RETRACT, E-PREDATORY, and W-UNVERIFIED, making it easier to aggregate results and route uncertain items for manual review.

Boundaries

It is useful for journal articles, conference papers, theses, books, patents, and web resources in Chinese or English, and can accept PDF, Word, OCR screenshots, or plain text. However, outcomes depend on database coverage, access rights, and page parsing. When no authoritative record is found, the skill usually marks the item as unverified or suspected false rather than declaring it definitively fake. For lists over 30 entries, it may prioritize suspicious items and sample the rest, so final judgments should still be checked against official pages.