GitHub

Working with ArXiv Papers

Transform academic research papers into interactive, explorable experiences.

Introduction

ArXiv is a repository of electronic preprints for scientific papers. Explorable Research can transform these papers into interactive visualizations that make complex research accessible to everyone.

How to Use ArXiv Papers

Step 1: Find the Paper

Go to arxiv.org and find the paper you want to visualize. Copy the URL from your browser.

Step 2: Paste the URL

Go to the Create page and paste the ArXiv URL in the input field. The system accepts multiple URL formats:

Supported URL formats
# Abstract page URL
https://arxiv.org/abs/2301.07067

# PDF URL
https://arxiv.org/pdf/2301.07067.pdf

# HTML version (extracts paper figures for richer context)
https://arxiv.org/html/2301.07067v1

# Just the arXiv ID
2301.07067

Step 3: Add Context (Optional)

You can add additional instructions to guide the AI on what aspects of the paper to focus on:

Adding context
https://arxiv.org/abs/2301.07067

Focus on the attention mechanism described in Section 3.
Include an interactive visualization of the self-attention weights.

Tips for Best Results

Focus on Key Concepts

Research papers often have many ideas. Specify which concept, figure, or algorithm you want to visualize for better results.

Reference Specific Sections

Point to specific sections, figures, or equations: “Visualize the architecture from Figure 2” or “Explain Equation 5 interactively.”

Simplify Complex Papers

For very complex papers, ask the AI to create a simplified explanation first, then add interactive elements.

Use HTML Links for Richer Context

When available, use the HTML version URL (e.g., arxiv.org/html/...) instead of the abstract or PDF link. HTML pages let us extract the paper's figures and diagrams, giving the AI visual context to create better explorables. Not all papers have an HTML version.

Iterate on Results

Use follow-up messages to refine the explorable. Ask for more detail on specific concepts or additional interactivity.

Example Use Cases

Transformer Architecture

Attention Is All You Need
https://arxiv.org/abs/1706.03762

Create an interactive visualization of the transformer architecture
showing how self-attention works. Include:
- Step-by-step attention calculation
- Query, Key, Value visualization
- Multi-head attention animation
- Position encoding explanation

Diffusion Models

Denoising Diffusion Probabilistic Models
https://arxiv.org/abs/2006.11239

Visualize the denoising process in diffusion models:
- Forward diffusion (adding noise step by step)
- Reverse diffusion (denoising)
- Interactive noise level slider
- Show how images are generated from noise

Graph Neural Networks

Semi-Supervised Classification with GCNs
https://arxiv.org/abs/1609.02907

Create an explorable about Graph Convolutional Networks:
- Interactive graph with draggable nodes
- Message passing visualization
- Feature aggregation animation
- Node classification demo

What Types of Papers Work Best?

While Explorable Research can handle any paper, certain types produce particularly good results:

  • Machine Learning papers – Neural network architectures, training algorithms, optimization methods
  • Computer Vision papers – Image processing, object detection, generative models
  • Physics papers – Simulations, mathematical models, physical systems
  • Algorithms papers – Data structures, graph algorithms, optimization
  • Mathematics papers – Geometric concepts, statistical methods, probability

Limitations

Keep these limitations in mind when working with ArXiv papers:

  • Very long papers may not be fully processed – focus on specific sections
  • Papers with complex mathematical notation may need additional guidance
  • Domain-specific jargon might need explanation for the AI
  • Papers requiring proprietary data cannot be fully reproduced

Next Steps

Try with a PaperSee All Features