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.
Presentations on other tracks at the same time
