Live Stream Inspection Agent/Custom script scenario
This tutorial shows you how to create a custom script scenario.
Overview
Use case: Your business needs to ensure that hosts follow specific scripts during live streams. This feature prevents hosts from using unapproved phrases, such as asking viewers for personal contact information or mentioning unauthorized return policies.
Feature overview: The script violation scenario feature lets you configure custom keywords and exclusion words. It uses AI-assisted recognition to improve the detection accuracy for your defined violation scenarios.
Procedure
Step 1: Create a scenario
Navigate to Live Stream Inspection Configuration > inspection management > inspection scenario management, and then click New Scenario.
Note: The product provides default, general-purpose inspection scenarios for live streams. These help you understand the enhanced accuracy of AI-powered inspection.
Enter the following information: Scenario Name, Scenario Description, keywords, and exclusion words (optional).
Keyword rules: An inspection alert is triggered if keywords are detected. Keywords are evaluated using OR logic within a group and AND logic between groups.
Exclusion word rules: An exclusion word prevents an inspection alert from being triggered. For example, if A is a keyword and B is an exclusion word, a sentence containing both A and B does not trigger a violation detection.
For example, you can configure a scenario as follows: If a host mentions any of the keywords in the first group (such as "world-leading" or "number one") AND also mentions the keyword in the second group ("our brand"), the system identifies it as a violation.
Step 2: Enable large model enhancement
You can enable the large model to automatically generate a prompt for AI-assisted recognition with a single click.
Prerequisites: Complete the Scenario Name, Scenario Description, keywords, and exclusion words (if any) fields. The system uses this information to generate the AI prompt. If you update the scenario details, regenerate the prompt.
Important: Do not modify the content that appears after ###Output.
Step 3: Debug the inspection scenario
You can use two methods to test the scenario: Historical data and Manual input.
Historical data: Select up to 10 records from the last 200 detected violations to test the scenario.
Manual input: Enter a sample script from a host to check if the large model-enhanced scenario correctly identifies it as a violation.
Historical data
Manual input
Step 4: Manage the scenario
Disable: Disables the scenario. The system stops using this scenario to inspect configured live streams.
Enable: Enables the scenario. The system resumes using this scenario to inspect configured live streams.
Delete: Permanently removes the scenario. This action is irreversible.
Step 5: Configure script inspection
Click Add Inspection or Edit to configure the inspection.