Set up and deploy detect intent models
Not sure what a detect intent model or matching phrase is? Read our tutorial on building a detect intent model before you start.
What you'll need
At least one intent
For each intent, at least three matching phrases
Step one: Set up your intent model
Go to Automate work > AI Hub > Detect intent
Click Create new detect intent model.
In the Model name field, enter a descriptive name for your model, such as Complaints.
In the Set up Intents section, click + Add new intent. A panel will open on the right-hand side of your screen.
Click + New matching phrase.
In the text field, enter a matching phrase.
Click + New matching phrase and add another phrase. Continue this process until you have added all of your matching phrase variations.
If you want to delete a matching phrase variation, click the ‘delete’ icon next to the question.
Click Save.
[Optional] In the ‘Irrelevant content’ section, click Irrelevant content. Click + New matching phrase and enter the irrelevant content that you want to exclude. Click Save.
If you have more than one intent that you would like to include in this detect intent model, repeat steps 4-10.
To publish your detect intent model and begin training the model, click Publish.
Success! You've just set up an intent model!
Step two: Train and test your model
Once you’ve published a detect intent model, you’ll have to wait for the model to be trained before you can use it. The training process happens automatically when you publish a detect intent model - you don’t need to do anything else.
Training a model is typically a quick process, but it will take longer if you have lots of detect intent, and/or lots of matching phrase examples.
You can see the status of your detect intent model by going to Automate work > AI Hub > Detect intent and looking under the ‘Status’ column.
When you first publish a detect intent model, its status will be ‘Training’.
If you edit and publish a detect intent model, its stats will be ‘Retraining’.
When your detect intent model is ready to use, its status will be ‘Ready’.
Test your model
Before you use your detect intent model, you’ll first need to test it to make sure that it works as expected.
Go to Automate work > AI Hub > Detect intent and select the model that you want to test.
Click View.
In the top right-hand corner, click Test model. A panel will open.
In the text field, type a question, then click Reply.
If the matching phrase you write is recognized and a matching answer is found, this will be returned here.
If the matching phrase you provided isn’t recognized or is matched to an irrelevant question that you’ve provided, this will be returned here.
The model will ask you if it has provided you with the correct intent for your matching phrase. If it has provided the correct intent, click Yes, then click Finish and train. The question will be added to the list of example matching phrases that you provided for that intent.
If it has provided an incorrect answer, click No. Select the correct answer from the drop-down, then click Finish and train. The question will be added to the list of example questions that you provided for that intent.
Step three: Deploy your model
Now that you’ve created your detect intent model, you can deploy this model using detect intent action for Flows.
Edit model
You can edit the model at any time by going to Automate work > AI Hub > Detect intent, selecting the model that you want to make changes to, and clicking Edit.
Make your changes, then click Publish to retrain your model.
Versioning
When you make changes to your model, we create a new version of that model and archive the previous version.
Go to Automate work > AI Hub > Detect intent and select the model you want to download.
Click View.
Click the Versions tab.
Download a model
Models can be downloaded as YAML (.yml) files.
Go to Automate work > AI Hub > Detect intent and select the model you want to download.
Click View.
Click the Actions button, then click Download.
Last updated