I’ve implemented a simple FAQ-bot w/ dozens of retrieval faq/ subintents and several chitchat intents. While it steadily recognizes individual intents, the real-life scenario is that users typically include multiple requests in one message. E.g., w/ the intents listed below, we can have "Hi! I couldn't find where to sign up. I'm looking for cheapest fridges"
# nlu.yml
- intent: faq/sign_up
examples: |
- can I sign up with my google account
- intent: faq/cheap_fridges
examples: |
- show me your cheapest refrigerators
I’m aware of this discussion and its predecessor but I’m afraid generating combinations of intents is infeasible. So I thought maybe I can catch users’ message, split it into sentences, and retreive intents for each sentence to return the corresponding responses. With our example above, I expect to recognize intents greet, faq/sign_up, and faq/cheap_fridges in only one message and render the corrseponding utter_ messages. Is it possible to implement within the default pipeline?