scientific-taxonomy

Scientific Taxonomy

Organize a supplied collection of scientific papers into one coherent topic hierarchy and write the result as Markdown.

Input readiness

Build the taxonomy from these fields only:

• paper_id: optional stable identifier used internally for deduplication and AMiner URL construction. Do not display it in the Markdown paper list.
• title: required for every paper.
• abstract: the complete abstract, required when it can be retrieved. An abstract_slice, snippet, or truncated preview is not a complete abstract.
• venue: conference or journal name, required when it can be retrieved.
• year: publication year, required when it can be retrieved.
• aminer_url: preferred paper link. If absent and a valid AMiner paper_id exists, construct https://www.aminer.cn/pub/{paper_id}.

Before classifying, check whether the supplied information is sufficient:

1. When the user supplies local PDFs instead of titles and abstracts, invoke an available local PDF-extraction tool or Skill first. Consume its structured title and abstract output; do not implement PDF parsing in this Skill.
2. Treat abstract_slice, search snippets, and visibly truncated abstracts as insufficient taxonomy input. For AMiner search results, invoke aminer-academic-search and use paper_detail with each selected AMiner paper ID to retrieve the complete abstract before classifying. Follow that Skill's authorization, pricing, and high-cost confirmation rules.
3. When a paper is supplied only as a DOI, arXiv ID, URL, or title, use an available AMiner Skill or AMiner API to resolve the paper and retrieve its complete abstract. Prefer a free lookup route for identity resolution, then use full paper details when the free result contains only an abstract slice.
4. Enrich only papers already supplied by the user. Do not add related papers unless the user explicitly asks.
5. Never invent or expand an abstract. If full-abstract retrieval is unavailable or returns no abstract, classify from the verified title and available slice, then record the affected paper titles in a short metadata note without visibly displaying paper IDs.