Retrieval intent is designed to mitigate the development complexity by grouping similar intents into one. However, there is a use case that relies on the prediction score on a specific sub-intent while the existing environment had little support to achieve this.
To specify, a set of retrieval intents are created for a FAQ bot. In order to facilitate the conversation, the bot would ask for rephrasing or clarification upon user messages with ambiguous intention. As a result, a detailed report of prediction scores within the set of retrieval intents is required. With the score, the bot would ask for an assurance of the most probable sub-intent and ask for correction if there is a mistaken prediction.
Use the following paragraphs for a longer description, or to establish category guidelines or rules:
- Why should people use this category? What is it for?
The prediction score of every sub-intent is used for making improvements to the fallback actions with use cases that retrieval intent is implemented.
- How exactly is this different than the other categories we already have?
The prediction score of sub-intents are undetermined in the existing environment
- What should topics in this category generally contain?
Fallback actions
- Do we need this category? Can we merge with another category or subcategory?
It can be added to the retrieval intent category in Rasa Open Source