Botfront is based on Rasa and provides interfaces to visually and efficiently build and edit Rasa stories. Stories are the building blocks of conversation flows. If you don’t know Rasa, it’s a framework using a simbolic language to design conversations.
You can visually create real conversation examples for your virtual assistant, with an experience very similar to a regular chat.
You can create your flows, branch them, link conversations to each other (or even with themselves depending on the situation).
You can annotate user utterances and add them to your training data, saving a lot of time while developing your model. Bot responses will also be automatically created and saved, taking away the need of filling them up with content elsewhere.
This is the initial screen when you create a new story in the conversation builder. Depending on the flow you want, it gives all the options to start from.
A story almost always starts with a user utterance (unless you are creating a destination story for linking).
Here we select the text option, payload will be shown in a later step:
The user says Hi there.
If you have a model (usually the case), the input would be parsed through and your model will annotate the utterance accordingly (as it would happen in New utterances in Incoming). If you don’t have a model yet, an intent won’t be attained and you would need to select/create one in order to save.
We can see existing intents to select from or create a new intent by clicking on the intent label.
We can make as many changes as we want for intents, annotate entities (if applicable) and save the user utterance.
You cannot edit a user utterance after you save it. If you want to change something, you will need to delete it and add a new one in the same place.
When you save a user utterance, it is also added to your training data to easily develop and enhance your model. If you delete it in the conversation builder, it will still remain in your training data. If you completely want to remove a user utterance from everywhere, you will need to manually delete it from your training data as well.
In our sample flow, we will continue with a bot response. Depending on your conversational needs, a slot or an action might come here as well. There cannot be two consecutive user utterances.
Here we continue with a text response:
We can type a simple text response in the bubble, just like chatting. The field would work with Markdown syntax as well.
We continue with a quick reply response this time.
Here we can set as many buttons as we want, which would be presented as clickable options to the user in the chat. They can have two types: Postback and Web URL:
- Postback would post the button payload back to the chat, as if the user “said” it. You just have to select the combination of intent and entities that you would like the bot to receive when a user clicks on the button.
- Web URL will open the URL put in the button settings.
Given that we are presenting two options to the user, it would be good to branch the conversation here and continue with two separate flows.
Here we continue with the payload option, in order to exactly match the payload we set for the button:
We can either create a new intent, or select an existing one.
We select the
Even though the payload option is selected, you might see content from an example that exists in your model, with the exact same payload. This will not affect the performance of the story or anything related with the buttons.
We do the same thing for the Sad branch.
And voilà! We have created a simple yet powerful story with branches. Now it’s time to train your model and try it out!
Here you can see an example of how you can use entities while annotating in the conversation builder:
Simply highlight the part (entity value) you want to annotate. From the popup, you may create a new entity, or select from an existing one in your model.
Depending on your conversation flow, you can use slots. Slots can be used in the beginning of branches. Here is a simple example:
If you have custom actions, you can use them in your conversation as follows:
Botfront exposes all the contextual power of Rasa. The context of a conversation is the knowledge of all the passed events of this conversation. In other words, you should be able to determine the response of a virtual assistant depending on various past events of the conversations, not just the last user utterance. The conversation below illustrates a very simple example of a context aware conversation. If a user repeats several times the exact same question, you might decide to treat that differently.
In the next section you will see other examples of context aware conversations, for example using slots to branch a conversation.