As in the title what are the main ideas behind stories generation using rasa? At some point, I understood that there is a graph generation step but I couldn’t relate it to anything.
Hey @dma092. In order to train a dialogue management model which will predict how the chatbot should respond to user inputs, you need to gather some conversations between a user and a bot and use them to train the model. A story is one of those conversations, converted in a Rasa format where user inputs are expressed as intents while the answers of the bot are expressed as actions. Rasa Core uses those stories to learn how users interact with the bot and use that knowledge to predict the responses of the bot in new conversations. You can read more about the stories format here and here.
HI @Juste. Have one more question regarding dialog training.
if training stories will be too detailed is there a chance to get Overfitted model?
Is there some guidance for it?
Thanks 
Hey @meksikann. It depends on what conversational turns those stories cover as well as how many training stories do you have. If the training stories you include do not drift much from the happy path then it’s likely that your model will overfit heavily. Detail in stories might also constribute to overfitting at the early stage of the bots development, but as your training data grows, the quality of the model should also increase. My advice would start with stories which help build a model that alows to handle the happy path easily and then build up the training data from there to include conversations which drift from it. Also, always run the evaluation scripts to see how the model is performing and inspect your data accordingly.
I appreciate the answer. can you provide us with more details about how the stories are augmented? (E.G: paper or other references)? Thank you in advance!
Hi @Juste thanks a lot for such simple and clear explanation !!