---

View the PDF Days Europe 2025 agenda

This presentaton is part of PDF Days Europe 2025.
Register now!

View our terms and conditions.

Tagged PDF in the Wild

Evaluating quality of real world Tagged PDFs

Session description

We analyse large volume of real world PDF coming from two public sources:

• Digital Corpora (8M PDFs)
• randomly crawled PDFs on the Web with recent creation dates

These collections are checked for the availability of tags and the syntax validity of the structure tree. In addition, all tagged PDFs with table structure elements are compared against AI-based table recognition. Extra details are given to the processing architecture, as analyzing 8M PDFs in a reasonable time and cost is a separate challenge.

---