We haven’t devoted much time to negative examples at Teneo.ai because their use is not one of the core concepts. Basically negative examples have the following function:
- Tested by autotest, to be sure they DO NOT trigger the associated condition. Described here: https://www.teneo.ai/studio/refine-your-bot/how-to/how-to-test-your-solution
- Used by NLU Generation (auto coding): if you generate a condition based on positive examples, it is assured that the condition is not true for the negative examples. See also: https://www.teneo.ai/studio/flows/how-to/create-a-syntax-trigger#generate-a-condition
- They are NOT used by class triggers, but can be a reference for you to create a hybrid trigger: one where you assure that the negative examples do not trigger the flow, even if the class is selected. See also: https://www.teneo.ai/studio/flows/concepts/triggers#class-triggers
As for what to use as a negative example: We usually wait until we see which inputs falsely trigger a flow to see what should be added. Don’t spend too much time worrying about all the inputs that should not enter the flow.
In general these would be inputs that look very similar to the positive examples but meaning something different. For example: something that triggers on “I love you” should not trigger on “I don’t love you”. If it does, add the negative example, and regenerate the condition.
Let me know please if you have any further questions