Synchronous Classification Endpoint

This endpoint allows you to classify text based on the labels you have set up synchronously. This endpoint analyzes the input text and returns predictions about which labels apply, including the confidence level for each prediction.

Endpoint Overview

The Classification endpoint is designed to classify text (e.g., emails, messages) into predefined categories (labels). The request can either include subject and metadata fields or just the body of a message. The response provides a list of detected labels, including the prediction confidence and priority.

Endpoint: POST /api/companies/{companyCuid}/orchestrators/{orchestratorCuid}/classification

Request Parameters

The Classification endpoint accepts the following parameters in the body of the request:

Payload Structure

Field	Type	Required	Description
subject	String	No	The subject or title of the text (optional).
metadata	Object	No	Any additional metadata related to the text (optional).
body	String	Yes	The body of the message or the main content of the text to classify.

Example Request Payload

{
  "subject": "Complaint about my order",
  "metadata": {
    "order_id": "12345",
    "customer_id": "67890"
  },
  "body": "I would like to file a complaint about my recent purchase. The item is damaged."
}

Response Structure

The response from the Classification endpoint will provide predictions, including a main prediction (the most probable label) and a list of predictions with their confidence and priority.

Response Fields

Field	Type	Description
main_prediction	String	The most likely label for the input text (e.g., "reclamation").
predictions	Array	An array of objects, each containing a label and its confidence.
predictions.label	String	The name of the predicted label.
predictions.confidence	String	The confidence level of the prediction (e.g., "high", "medium", "low").
predictions.priority	Integer	A priority score to indicate the relevance of the prediction.

Example Response Payload

{
  "main_prediction": "reclamation",
  "predictions": [
    {
      "label": "reclamation",
      "confidence": "high",
      "priority": 1
    },
    {
      "label": "product_issue",
      "confidence": "medium",
      "priority": 2
    }
  ]
}

Best Practices for Using the Classification Endpoint

Provide Context: If possible, always include the subject and metadata in the request to give the model more context for better predictions.
Use Clear Labels: Ensure your labels are clearly defined and represent distinct categories to avoid confusion in the predictions.

For additional information on managing labels or examples, refer to the following pages:

Endpoint Overview​

Request Parameters​

Payload Structure​

Example Request Payload​

Response Structure​

Response Fields​

Example Response Payload​

Best Practices for Using the Classification Endpoint​