The Future of Documents

One exciting piece of  news that will make a difference in the Digital Mailroom, but more generally in any Business Process where paper documents or dossiers need to be classified according to complex rules, is Text and Image categorization (also called “Multimodal”). 

Text and Image Categorization combines both text and image information to learn complex business rules, where both textual content and image content are required to identify a document category.

Let’s take as example an insurance claim. A dossier, consisting of multiple documents, is received by an insurance company, and might contain a claim letter and photos of the damage.

Even this simple case is a challenge for textual or image-based classification. Indeed, image recognition might find a letter can be representative of any type correspondence, while a photo might be contained in multiple types of reports. Text categorization alone might be able to tell from the words that it is related to a claim, but might not be able to tell whether it’s a house incident, a car accident, or something else.

Multimodal categorization will be able to combine the detection of a typed letter, the extraction of its characteristic vocabulary (e.g. characteristic of an insurance claim), with the presence of a photo, or even better a type of photo (e.g. a wrecked car) to uniquely identify the dossier as an automobile claim.