Split by Intent: Using Classification Artificial Intelligence

Mallu Updated by Mallu

What are Classification AIs (Classification Intelligences) or Classifiers?

Classifiers refer to platforms and artificial intelligence systems responsible for categorizing responses and inputs from contacts, returning keywords that can be used in decision card rules.

Weni Platform has its own repository of intelligences, which can be created from scratch or by using public intelligences.

To access it, simply click on the Artificial Intelligence Module and select Content Classification from the icon menu on the left side of the screen.

In proprietary intelligences, you will be able to view the AIs that have already been integrated into your project. To add a new one, click on Public Intelligences, choose the desired classification AI, and integrate it into the project by clicking the + icon.

Now intelligence is available to be used within flows!

Flow Anatomy

The intelligence used within flows is based on classifications, where the user will type a phrase or keyword, and this input will be analyzed by a "Wait for Response" card in the Has a text function. This card allows all user responses to be processed by the intelligence.

The inputted text will be sent to the intelligence as @input.text, which will analyze and return intentions, meaning a phrase or word that tells us what the message is about. Additionally, it returns a confidence score, indicating how certain the classification AI is that the result is correct.

Split by Intent Card

  1. In the Split by Intent card, you will choose the classification AI (available as explained above) to be used from the list under the phrase "Run the last response through the classifier."
  2. After that, you can choose the intentions that will be used:
    The Split by Intent card is a special decision card that only appears in the list when a classification AI has been created. This decision card has a single rule: it checks the intent and confidence level to define a route.
    For example: If the returned intent is Y and the classifier has more than 80% confidence, direct it to category X.
    Below is the anatomy of the card:

Example: If the returned intent is negative and the classifier has more than 80% confidence, direct it to the Negative category.

By default, the "Split by Intent" card analyzes the last message sent by the user. However, you can change the value to be analyzed in the Classifier Input tab, where it is possible to add any valid platform expression or variable that returns a text value.

How did we do?

Call Webhook: Making requests to external services

Import and export flows

Contact