Enable subtree classification for a class

Use subtree classification to help differentiate a parent class from its child classes. The relationship between parent and child classes can be as simple as different versions of the same document, but this is not always the case. When classification is run, documents belonging to the parent or its child classes are initially classified as the parent. Subtree classification is performed to determine whether the document belongs a child class instead of the parent class.

Note It is common for projects that use subtree classification to exclude the parent class as a valid classification result. In many cases, the parent class is collector or category rather than a document type, and the child classes are the actual classification results. During classification, the parent class is not a valid classification result, so a child class is assigned, even if it has a lower confidence than the parent class.

Before using subtree classification, you must enable subtree classification for the parent class by following these steps:

  1. In the main Transformation Designer window, hide or close any windows that block the Ribbon.
  2. Open the Project Tree window if it is not already open.
  3. Expand the Project Tree and select the class.
  4. Optionally, view the class contents if they are not already displayed.

    The hidden class contents are displayed.

  5. Open the Details window if it is not already open.
  6. On the Class Details window, in the Subtree Classification group, select Enable subtree classification.

    The subtree classification options are enabled.

  7. Optionally, adjust the Minimum confidence option.

    This value determines the confidence level needed to classify a document as belonging to the child class.

  8. Optionally, adjust the Minimum distance option.

    This value sets the confidence difference that is required between the highest confidence and the next highest confidence, for the highest confidence to be selected as the classification result.

  9. Add training documents for each child class.
  10. Train your project.

    Subtree classifiers are created automatically in the class properties of each parent class with subtree classification enabled.