I am creating a device troubleshooting chatbot. I have around 50+ stories for a single device covering all possible conversation flows with some repetitive stories as there are around 10 different flows. I trained rasa_core on the given 50+ sample stories and when i test the bot, it always predict wrong action and gives wrong utter message except for the greet intent.
I have tried:
- Memoization, Keras and Fallback Policies
- Keras and Fallback Policies
- Custom Keras Policy But always get wrong result for action prediction.
Anyone can help me out?