Powered by Zoomin Software. For more details please contactZoomin

The Semaphore Fact Extraction Framework (FACTS)

Types document template

  • Last Updated: July 8, 2026
  • 2 minute read
    • Semaphore
    • Documentation

Fact Extraction Framework Document Types

Representing Document Types brings two main benefits:

  1. It provides a way to identify which documents should have their facts extracted by which extractors, increasing fact extraction performance.
  2. It provides an organizing principle in the model for the facts and their extractors.

Often, extraction is based around particular document templates in an organization. One common way to organize contexts is by document template type. Typically, you would model one Document Type for each document template. All the facts you wish to extract for any document template should be modeled as belonging to that Document Type.

A key question then arises: how do you know that you are looking at a particular document template at extraction time? The FACTS framework addresses this by representing a “fingerprint” that identifies the document as being of a certain Document Type. Extracting facts from their proper context—the right document—greatly increases the performance of fact extraction.

However, sometimes the facts you wish to extract are not attributable to a single document template; the facts may appear in any or all templates, regardless of the particular document template. In that case, you can define an abstract document type that collates all the facts you wish to extract, without independent fingerprinting. Essentially, you are saying: if the document has this fact, then it is of this type; the facts self-identify.

Whether the fact self-identifies or is independently fingerprinted, both are managed through Document Facts. Document Fact is the next level down from Document Type. To represent which facts belong to which document type, we use Document Facts. Document Types can have one or more Document Facts.

Types of Content

A document type can be associated with one or more Content Types. A content type is a logical abstraction away from the document type. Whereas document types require structural similarity to apply the correct extractors, a content type requires informational similarity in the content. For example, you may have several press release document templates, but they are all of type Press Release (the content type).

TitleResults for “How to create a CRG?”Also Available inAlert