Configure notifications

Key concepts

Before you configure callback notifications, understand the following concepts.

Scan result callback notification

All asynchronous AI Guardrails API operations support scan result callbacks, such as asynchronous image scan and asynchronous video scan. To receive scan results via callback, you must include the callback (webhook address) and seed (verification string) parameters in your asynchronous API request.

Procedure

1. Prepare an HTTP webhook address and a seed parameter to receive scan results.

2. When you call an asynchronous AI Guardrails API operation, pass the corresponding callback and seed request parameters. For more information about the parameters, see the parameter description in the API reference for the specific operation.

Human review callback notification

Human review API operations do not return detection results in real time. Instead, the system notifies the caller by using callbacks. Human review falls into two categories: machine-assisted human review and pure human review. The callback method differs for each category, as described in the following sections.

• Machine-assisted human review

  If you use the Alibaba Cloud human review service together with the machine moderation service, you must configure notifications on the console to receive callbacks for the human review module.

  1. Log on to the AI Guardrails console.

  2. In the left-side navigation pane, choose Machine Moderation V1.0 > Settings.

  3. On the Machine Moderation page, click the Notification tab, and then click Create New Notification.

  4. In the Create New Notification dialog box, enter a Title and Callback URL, and select an Encryption algorithm, a Notification type, and a Moderation Result. Then, click OK.

     After you save the configuration, the system automatically generates a seed. The seed verifies that requests to your callback endpoint are from Alibaba Cloud. Save the automatically generated seed.

     Important

     • If you have already configured a callback notification service for machine moderation, you can reuse the existing configuration or create a new one based on your business requirements.

     • You can receive callbacks for all review results, or select specific results based on your business scenarios.

     • For Manual Review Results by AlibabaCloud, select Alibaba Cloud Human Review Result.

     • Encryption algorithms:

       • SHA256: Uses the HMAC-SHA256 encryption algorithm.

       • SM3: Uses the SM3 encryption algorithm. It returns a lowercase hexadecimal string.

         For example, encrypting abc with SM3 returns 66c7f0f462eeedd9d1f2d46bdc10e4e24167c4875cf2f7a2297da02b8f4ba8e0.

  5. On the BizType Manage tab, find the target business scenario, click Choose Notification Plan in the Actions column, and associate the Callback notification scheme you configured.

• Pure human review

  If you use the Alibaba Cloud pure human review service, you must include the callback (webhook address) and seed (verification string) parameters in your request to receive human review callbacks.

  1. Prepare an HTTP webhook address and a seed parameter to receive scan results.

  2. When you call an asynchronous AI Guardrails API operation, pass the corresponding callback and seed request parameters.

     For more information about the parameters, see Human review for images, Human review for videos, Human review for text, and Review human review results.

Content field

After you enable callback notifications, AI Guardrails sends notifications that include a content form field. The following table describes the structure of this field.

{
    "scanResult": {
        "code": 200,
        "msg": "OK",
        "taskId": "fdd25f95-4892-4d6b-aca9-7939bc6e9baa-1486198766695",
        "url": "http://1.jpg",
        "results": [
            {
                "rate": 100,
                "scene": "porn",
                "suggestion": "block",
                "label": "porn"
            }
        ]
    },
    "auditResult": {
        "suggestion": "block",
        "labels": [
            "porn",
            "ad",
            "terrorism"
        ]
    },
    "humanAuditResult": {
        "suggestion": "pass",
          "dataId": "yyyy",
          "labels": [
              "porn",
              "vulgar"
        ],
        "taskId": "xxxxxx"
    }
}