Multi-turn conversations using TED Policy

Rasa Open Source
1

I recently came across the tutorial on TED Policy where where it handles non-linear conversations such as when the user interjects with an off-topic message or loops back to modify an earlier statement.

The blog post provides a sample conversation.

TED%20-%20more%20layers
TED%20-%20more%20layers960×720 125 KB

How do we prepare the training dataset (stories) to train the model like in the image using TED ?

Will the bot automatically get trained to handle the multi-turns and non-linearity if we have a happy path and just include TEDPolicy in our config.yml file? Or if we need to make changes to the dataset, how to do so?

Thanks.

2

Saugat,

There are no special requirements for to use TED. Write stories as described in the docs and give it a try.

Greg

3

Hi Greg.

Thanks for your reply. It would have been great if Rasa team would build a sample bot demonstrating the strength of TED Policy handling these kind of non-linearity and multi-turns.

Writing stories to cover all the possible paths is pretty tiresome. And hey, where is the use of machine learning if it can’t handle these scenarios.

4

You should not need to write stories to cover all possible paths. Start with three stories, happy and unhappy path stories and another example with an out of context intent that returns to the story.

## restaurant with out of context chitchat
* request_restaurant
    - restaurant_form
    - form{"name": "restaurant_form"}
* chitchat
    - utter_chitchat
    - restaurant_form
    - form{"name": null}