With Rasa X 0.27.0 and 0.28.0, we’ve released improvements to the NLU Inbox that make it easier to review and annotate incoming messages.
Here’s what’s new:
- The NLU Inbox has been relocated to the side menu, making it more discoverable
- We’ve added an indicator showing the number of unreviewed messages
- You can now use keyboard shortcuts to navigate the Inbox
The NLU Inbox collects messages users send to your assistant and allows you to annotate and convert those messages into new training data. Over time, enriching your training data with real user messages results in assistants that can handle the real-life challenges users throw at them. With these recent changes to the NLU Inbox, along with backend performance enhancements, Rasa X now provides a more intuitive and efficient experience for annotating data.
The NLU Inbox Workflow
When users talk to your assistant—via a messaging channel, the Share your bot feature, or through the Talk to your bot screen—their messages are funneled into the NLU inbox. When you have unprocessed messages in the Inbox, you’ll now see an indicator in the sidebar, alerting you that messages are ready to be reviewed.
For each user message in the list, you’ll see which intent was predicted for that message, along with the model’s confidence. You’ll also see which entities were extracted from the message, highlighted in orange. If the model’s classifications were correct, you can click the checkmark button to save the message to your NLU data file. If the predictions weren’t correct, you can adjust the intent and entity labels (by assigning a different label or creating a new one). Then, you can save the message to your training data. You can also remove outliers using the NLU Inbox. If a message comes in that you don’t want to add to your training data, simply delete it.
Now, we’ve enhanced this workflow with the addition of keyboard shortcuts. The J key moves up in the list, and the K key moves down. You can save a message by hitting the S key, or delete it using D.
Rasa X is the tool for teams and developers who follow Conversation-Driven Development—that is, use insights from users to improve their AI assistant. One of the core principles of CDD is building a training data set that reflects the way users really talk to your assistant. When going into production, more than 90% of your training data should originate from real messages with users, and the NLU Inbox helps you build that data set.
We’re always working to make the NLU Inbox and Rasa X a tool that teams love to use, based on user feedback. Let us know how you’re using Rasa X to improve your assistant, and stay tuned for more updates!