Classification instructions
Issue
When setting up a classification project, you can add explicit instructions to classes.
Instructions are rules that the Instruction Classifier uses to analyze content to determine how to classify a document. If the document content contains phrases provided in a classification instruction, the confidence level of the classification is higher than if no phrases are found.
Solution
The logic to determine precedence of instructions is complex and some combination of instructions might not yield the expected results for a user not familiarised with how it works.
For example:
You have three words A, B and C. If you have one phrase "A B C", then you expect spatial proximity, while three phrases "A", "B" and "C" do not have to be spatially related. Thus, "A B", "C" wins over "A", "B", "C".
The recommendation would therefore be to use the Adaptive Feature Classifier, which provides trainable content classification, and add explicit instructions to classes to handle exceptions instead of relying just in instructions for the classification by adding lots of different instructions and phrases per class.
Level of Complexity
Moderate
Applies to
| Product | Version | Build | Environment | Hardware |
|---|---|---|---|---|
| KTA | 7.7+ |
References
Add any references to other internal or external articles