Paper Card Generator

Paper Card Generator

Overview

Create concise, evidence-grounded Chinese paper cards for academic reading, citation organization, related-work writing, and method comparison. Prefer official sources and preserve uncertainty instead of inventing missing experimental numbers.

Workflow

1. Identify the paper from the user input.

   • Accept title, DOI, official page, PDF link, arXiv page, project page, or a batch list.
   • If multiple papers match the same title, disambiguate with venue, year, authors, or URL before writing the card.

2. Gather primary evidence.

   • Use official PDF, conference/journal page, DOI page, arXiv page, and official code/project pages first.
   • For current or hard-to-verify details, browse and cite the sources used.
   • Read references/evidence-policy.md when experiment values, citation metadata, or publication status are uncertain.

3. Choose the card shape.

   • Use references/card-template.md for the default card format.
   • For one paper, output one complete card.
   • For batches, output one card per paper and start every card with ## Paper Title.

4. Extract the core content.

   • Capture title, method name, venue/year, research task, method type, backbone/base model/core model, official links, main idea, experimental setup, main results, and citation format.
   • Keep the summary focused on what the method does, how experiments are set up, and why it matters for comparison or citation.

5. Handle missing or unverifiable information explicitly.

   • Use the standard missing-information phrases from references/evidence-policy.md.
   • Mark arXiv-only papers as preprints and include version/date context when it matters.

Optional Comparison Fields

Add these fields only when the user asks for baseline selection, SOTA comparison, related-work grouping, or comparison against a specific method:

• Input modality: include only for multimodal, speech, vision, or modality-comparison tasks.
• Main-table recommendation: include only for baseline/SOTA table selection.
• Comparison with my method: include only when the user provides their method or asks for comparison.
• Suggested placement: include only for related-work or experiment-table grouping.

Do not include these fields in the default generic paper card.

Output Rules

• Write in Chinese by default, preserving necessary English terms, method names, datasets, and metric abbreviations.
• Do not output BibTeX unless the user explicitly asks for it.
• Use the user's provided citation style when available; otherwise use a numbered prose reference style with DOI as a Markdown link when available.
• For classification experiments, default to columns: Dataset / Setting / Classes or Labels / Metrics / Main Result.
• For generation, retrieval, regression, or open-ended evaluation, adapt the experiment table to the paper's original metrics instead of forcing a classification template.
• If asked to save results, update only the user-specified file. Otherwise, return the card in the conversation.