Search

Overview of Document Processing

Articles

Document Processing is the conversion of paper-based and electronic documents into digital information using the combination of Intelligent Character Recognition (ICR), Optical Character Recognition (OCR), Machine Learning (ML) algorithm, and necessary manual interventions.

Types of Documents

The types of documents and the nodes which are used to process them are listed below.

  • Documents in PDF format:
    • Use Doc Reader node to process Structured and Semi-Structured documents.
    • Use Hand Written Text Extraction node to process Handwritten Documents and pass it through the Doc Reader node to extract the data.
    • JIFFY.ai is not handling unstructured documents currently.
  • Use Excel node to process documents in Excel format.

    If the document contains image, install ABBY Fine Reader to convert image to editable text and pass it through the Doc Reader node to extract the data.

Out of the Box Capabilities

In JIFFY.ai, Invoice and Bill of Lading are provided as predefined schemas for ease of use. Invoice schema comes with thirty-five predefined fields and Bill of Lading schema with twelve predefined fields. Jiffy.ai automatically extracts information from these documents without any training and provides out-of-the-box machine learning models for these document types.

The model is already trained for the predefined schemas. When an Invoice or Bill of Lading is processed through the Doc Reader node, you do not need to train the ML. The data is extracted automatically from the documents using the built-in extraction modules.

For other documents, you may have to train using the point and click familiarization environment provided.

How is Document Processing Done in JIFFY.ai?

Document processing is achieved in four phases:

  • Create a Document Table with the required columns for the fields being extracted from the document. Document Table is the persistence layer to store, track and present extracted contents of the document being processed.
  • Design the task using the Doc Reader node to extract the fields from the document.
  • Execute the task to:
    • Categorize the documents: classify the document type and identify the classification group that the document falls in, based on the format of the document.
    • Populate the data into Document Table to store, track and present extracted contents of the document being processed. If Document Table is created using predefined schema, ML auto-extracts the required data and assigns a category based on the template of the document.
  • Familiarize the document: A user-friendly interface is provided to:
    • Point the labels and data to be extracted from the document, thereby training the model for the category of document being processed.
    • Verify and approve the fields extracted by the model.

If Document Table is created using custom schema, the fields are auto-extracted based on the existing trained model.

In an Invoice Processing HyperApp:

  1. A Document Table with name InvoiceTable is created using Invoice schema.
  2. A Task is designed with Doc Reader node to extract the fields from the Invoice.
  3. The Task is executed to extract the fields.
The document is familiarized, saved, and approved to train the ML engine for the category of document being processed. The approved fields are populated into InvoiceTable for further processing.

Did you find what you were looking for?