A company is converting many paper receipts into images and needs to extract entities like date, location, notes, and custom fields (receipt numbers). They already use OCR to get text but document layouts vary and building label workflows is time-consuming. They also have a small NER dataset that needs much more training data. Which approach will require the least effort to get reliable entity extraction?

This approach leverages fully managed AWS services designed for this exact use case, minimizing effort. Amazon Textract is ideal for extracting text and structured data from documents like receipts, handling varying layouts automatically. Amazon Comprehend offers pre-trained entity recognition for common entities (date, location) and Comprehend Custom Entity Recognition allows training models for specific custom fields (receipt numbers) with less data and effort than building a model from scratch. Incorrect options: Training a BlazingText model for NER is more complex and time-consuming than using Comprehend, especially for custom entities, and requires more data. Using a third-party OCR model adds integration complexity and potential cost without offering significant advantages over Textract for receipt processing. Pretrained NER deep-learning models might not cover custom fields well, necessitating more effort for customization. While using Textract for OCR is good, relying on a third-party NER model introduces similar integration and customization challenges as the previous incorrect option.