What is the best way to process tables in PDFs?

Transforming tables from PDFs can be quite challenging since most PDFs don't actually contain 'tables' but just text arranged in a certain way. Textract uses layout and machine learning to infer structure, which can lead to mixed results. I've had success with the stack I use:

There's a library called DsPdf which offers two methods: 1) layout-based extraction where you define a specific area to get tables, and 2) AI-based extraction where you describe the table you need. The first method works great for structured documents like invoices, while the AI-assisted method allows for more flexibility, especially with varying layouts.

Textract is functional but may be more than you need unless you're heavily invested in the AWS ecosystem. If your PDFs are digitally generated (not scanned), have a look at Tabula or pdfplumber. They often produce cleaner table outputs with much less hassle. For scanned documents, Google Document AI and Azure Form Recognizer tend to outperform Textract, especially in retaining the table format. In general:

- For Digital PDFs: Use Tabula or pdfplumber.
- For Scanned PDFs: Opt for Google Document AI or Azure.
- For maximum automation: Consider Textract or Document AI.

If simplicity and versatility are what you're after, I'd recommend DigiParser. It's designed to efficiently handle complex layouts in various PDF documents and can extract table data effectively.

Dealing with PDFs can be a hassle given their unstructured nature. However, I found Docling (a Python tool) to be incredibly helpful. It simplifies messy PDFs, DOCX files, and slides into structured data for various applications. It handles complex layouts and tables quite smoothly. Check out their site for more info!

Textract is solid for AWS users, but for alternatives, have a look at pdf.js for client-side extraction or pypdf2/pdfplumber for server-side in Python. For specific table extraction, tabula-py works wonders. If you need a more robust solution, Azure's Form Recognizer or Google Document AI are excellent choices. Just keep in mind that Textract can get pricey at scale, but it usually delivers good accuracy.