Academic Paper Analyzer - In-depth Analysis of Academic Papers

Core Capabilities

• MinerU Cloud API High-precision PDF Parsing
• Automatically extract images, tables, and LaTeX formulas
• Multiple Writing Styles: Storytelling / Academic / Concise
• Optional Formula Explanations: Insert formula images and provide detailed interpretations
• Optional Code Analysis: Explain in combination with GitHub open-source code
• Output Markdown + HTML (base64 image embedding)

Preparations

MinerU API Token

1. Visit https://mineru.net to register an account
2. Obtain the API Token
3. Set environment variables (recommended):
   bash

   export MINERU_TOKEN="your_token_here"

Dependency Installation

pip install requests markdown

Operation Steps

Step 1: PDF Parsing (Using MinerU API)

python scripts/mineru_api.py <pdf_path> <output_dir>

python scripts/mineru_api.py paper.pdf ./output YOUR_TOKEN

• output_dir/*.md

  - Markdown files (including formulas, tables)

• output_dir/images/

  - High-quality extracted images

Step 2: Extract Paper Information

python scripts/extract_paper_info.py <output_dir>/*.md paper_info.json

Step 3: Style Selection (Ask User)

1. Writing Style (Required)

2. Formula Options (Optional)

3. Code Options (Optional, only if the paper has GitHub links)

Please select an article style:

1. academic - Academic, professional and rigorous (recommended default)
2. storytelling - Narrative, down-to-earth
3. concise - Concise, quick reading

Do you need formula explanations? (Recommended if the paper contains mathematical formulas) Do you need to combine with GitHub code analysis? (Open-source repository detected: xxx)

Step 4: Intelligent Article Generation

• styles/storytelling.md

  - Storytelling Style Guide

• styles/academic.md

  - Academic Style Guide

• styles/concise.md

  - Concise Style Guide

• styles/with-formulas.md

  - Formula Explanation Guide

• styles/with-code.md

  - Code Analysis Guide

Lightweight Mode (Save Context)

1. Do not repeatedly read image files - MinerU has extracted high-quality images, just reference the path directly
2. Trust paper_info.json - Contains image lists and metadata, no visual confirmation needed
3. Only view key images - Read at most 1-2 core architecture diagrams, reference the rest directly
4. Let users verify - After generating HTML, let users check if the images are correct themselves

General Writing Principles

• Overused AI phrases (e.g., "in-depth discussion", "of vital importance", "in the field of...")
• Mechanized section titles
• LaTeX formula syntax (e.g.,

  $\mathcal{O}(1)$

  ) - Use extracted formula images instead
• Flat, straightforward technical descriptions

• Natural paragraph narration
• Make full use of images extracted by MinerU
• Every key figure in the paper should be explained
• Formula screenshots are more readable than LaTeX syntax

Storytelling Style Methodology (Exclusive for Storytelling Style)

• Wrong: "This paper proposes a conditional memory module based on hash tables"
• Correct: "Have you ever thought that large models actually don't have memory functions?"

• Before introducing new technologies, explain related old technologies first
• Let readers understand "why this innovation is needed"

• Choose a simple example and reuse it repeatedly
• For example: "The capital of China is Beijing"

• "Using a cannon to kill a mosquito", "Looking up in a dictionary vs memorizing a dictionary"
• Make abstract concepts concrete

• Simple problems → Complex problems → Solutions

• Summarize in one sentence, e.g., "Memory is memory, computation is computation"

Article Structure

**Paper Title**: xxx
**Paper Link**: [arXiv](https://arxiv.org/abs/xxxx)
**Author Team**: xxx

• Start with a question or scenario
• Arouse readers' curiosity
• Lead to "why this research is needed"

• Explain relevant basic technologies or historical methods
• Illustrate with simple examples
• Let readers understand the limitations of existing solutions

• Explain the paper's innovations in detail
• Each innovation should be supported by images
• Use metaphors and examples to make abstract concepts concrete
• Display formulas with images instead of LaTeX syntax

• Key experimental result charts
• Comparative analysis and data interpretation
• Highlight the most outstanding results

• Mechanism analysis, ablation experiments, etc.
• Explain "why this method works"
• Provide a deeper understanding

• Refine core insights
• Predict future development directions
• Personal opinions and evaluations

Step 5: Output Format (Ask User)

"Article generated:

article.md

. Do you need to generate an HTML version? (HTML embeds images for easy direct sharing)"

python scripts/generate_html.py <article.md> <output.html>

Resource Index

• styles/storytelling.md

  - Storytelling Style

• styles/academic.md

  - Academic Style

• styles/concise.md

  - Concise Style

• styles/with-formulas.md

  - Formula Explanations

• styles/with-code.md

  - Code Analysis

• scripts/mineru_api.py

  - MinerU Cloud API Call (Recommended)

• scripts/convert_pdf.py

  - Local Conversion (Alternative, requires PyMuPDF)

• scripts/extract_paper_info.py

  - Extract Paper Metadata

• scripts/generate_html.py

  - Generate HTML (base64 images)

Notes

• Prioritize using MinerU API for the highest accuracy, supporting formulas/tables
• Save context: Do not repeatedly read images, trust metadata
• Do not output the analysis process; users only need the final article
• Avoid bullet points, use natural paragraph narration
• Select 3-5 key charts as images

API Limitations

• Maximum size per file: 200MB
• Maximum pages per file: 600 pages
• Supports formats such as PDF, DOC, PPT, images, etc.