Azure Document Intelligence fails to accurately extract scientific notation

Hongqian Li 65 Reputation points
2025-07-17T18:18:15.9366667+00:00

We're using Azure Document Intelligence's prebuilt layout model to parse documents, but it's not accurately preserving scientific notation—e.g., 2.0 × 10⁻⁵ is extracted as 2.0 X 10-5. Subscripts and superscripts are not handled properly. Is there a way to configure the model or work around this to retain precise mathematical representations?

Azure Document Intelligence in Foundry Tools
0 comments

Answer accepted by question author

Chakaravarthi Rangarajan Bhargavi 1,280 Reputation points MVP
2025-07-18T03:15:20.3033333+00:00

Hi Hongqian Li,

Welcome to Microsoft Q&A!

Thank you for reaching out with this important question.

You're right — currently, the Prebuilt Layout Model in Azure Document Intelligence tends to simplify certain text patterns, especially when it comes to scientific notations and formatting details like superscripts or symbols. So, notations like 2.0 × 10⁻⁵ often get extracted as 2.0 X 10-5. This happens because the model is optimized for layout and structure rather than preserving mathematical fidelity or rich character formatting.

If you're working with scientific documents and precision is critical, here are a few directions that could help:

One practical option is to explore building a custom model using Document Intelligence Studio. You can upload your sample documents, label them carefully, and use your own post-processing logic (like regex) to correct these notations after extraction. This is particularly helpful if your documents follow a consistent structure.

Another approach is to combine Azure Document Intelligence with the Azure Vision API (Read OCR v4). The Read API can give you access to line positions, text spans, and font positioning. This data can be used to infer superscripts or symbols more accurately, especially when coupled with a custom parser.

If your use case involves heavy mathematical or scientific content, you might also look into specialized OCR tools like MathPix or LaTeX-focused OCR engines. These tools are designed specifically to handle formulas and symbols, and their output can be integrated into your pipeline alongside Document Intelligence for best results.

There’s no current configuration setting within the prebuilt layout model to preserve exact mathematical notation, but combining Azure’s capabilities with a bit of customization should help bridge the gap.

Here are a few Microsoft resources that might be helpful in this context:

Transparency Note – Azure Document Intelligence

Prebuilt Mortgage Model

Prebuilt Contract Model

AI Builder Integration with Azure Document Intelligence

If you find this answer helpful, please accept it and mark it as helpful. Do reach out again if you need further assistance with your implementation!

Warm regards,

Chakravarthi Rangarajan Bhargavi

Was this answer helpful?

1 person found this answer helpful.

0 additional answers

Sort by: Most helpful
Most helpful Newest Oldest

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.