The fallback mechanism in Rasa

In real life, there will always be situations that chatbots cannot handle. For example, the user input voice is not clear enough, or the requested service is beyond what the system can offer. Then we need a fallback operation to handle those exceptions so that we can still elegantly reply to users with something like Sorry, I could not understand what you meant. Categorized by triggering cause, fallbacks can be NLU fallback or policy fallback.

Now, let's start with NLU fallback.

Handling fallback in NLU

NLU fallback is used to handle situations where the NLU module cannot clearly understand what user's intent is. The FallbackClassifier component is used for this purpose, and its configuration example is as follows:

pipeline:
  - name: FallbackClassifier
    threshold: 0.6
    ambiguity_threshold: 0.1

Here, if the confidence of the intent with the highest score is equal to or lower...