Skip to content

Document splitting in Bedrock Data Automation

0

I'm trying to use Bedrock Data Automation for IDP purpose. But I have been observing some problems with document splitting. For multipage pdf file (one whole document), it will split the document when it is not suppose to, and for multipage pdf file that contains multiple individual files (For example, multiple receipts, where receipts has the exact same format [from the same shop]) it is not splitting how it should. In later case, only one receipt will be split and other receipts will be clustered still and extraction will go wrong. Is this because it is still new service? Or is there possible solution to this.

1 Answer
0

Amazon Bedrock Data Automation (BDA) does offer document splitting capabilities, but it's important to understand how it works and its limitations to address the issues you're experiencing.

For single documents that are being split when they shouldn't be: BDA's automatic splitting is designed to divide documents based on semantic boundaries. It supports files with up to 1000 pages and individual documents of up to 20 pages each. If your single document is being split incorrectly, it might be because the system is detecting what it perceives as semantic boundaries within the document.

For multiple documents in a single PDF that aren't splitting correctly: The splitting feature is designed to separate logical documents within a single file. However, its effectiveness can vary depending on the structure and content of the documents. In your case with multiple receipts, if only one receipt is being split out while others remain clustered, it suggests that the system is having difficulty identifying the boundaries between the receipts.

To address these issues, consider the following:

  1. Ensure that document splitting is enabled in your project configuration. This is done by setting the 'splitter' state to 'ENABLED' in the overrideConfiguration when creating or updating your project.

  2. Review your blueprint configurations. BDA can have up to 40 document blueprints attached to a single project. Make sure you have appropriate blueprints for each document type you're processing (e.g., a specific blueprint for receipts).

  3. Improve your blueprints. The accuracy of document splitting and classification depends on the quality of your blueprints and the consistency of your document formats. You may need to refine your blueprints over time to improve accuracy.

  4. For documents that shouldn't be split (like your single multi-page document), you might consider pre-processing these files to ensure they're treated as single documents.

  5. For multiple documents in a single file (like your receipts), if the automatic splitting isn't working effectively, you might need to pre-process these files to separate them before sending them to BDA.

Remember that BDA is a relatively new service and may still be evolving. If you continue to experience issues after trying these solutions, it would be advisable to contact AWS support for further assistance. They may be able to provide more specific guidance or inform you of any known limitations or upcoming improvements to the service.

Sources
AWS Bedrock to split document based on document type | AWS re:Post
Splitting documents while using projects - Amazon Bedrock
Bedrock Data Automation projects - Amazon Bedrock

answered 2 years ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.