Training Data Quality Management

The Training Data Quality Management page automatically scans registered knowledge to ensure that sensitive information is not included in the training data.
This feature is designed to enhance security and stability, especially in environments where it is unrealistic for operators to manually review every piece of uploaded content.

Review Options for Training Data

“Review Required” on Sensitive Information Detection

When this option is enabled, the system continues training even if sensitive information is detected.

However, all detected items are automatically added to the “Review Required” list so that administrators can review them afterward.

“Training Failure” on Sensitive Information Detection

When this option is selected, training is automatically stopped for any knowledge containing detected sensitive information.

This option is recommended for agents that handle highly sensitive or confidential data, providing a stricter level of data governance.

Review Pattern List

Pattern Name

Clearly identify the type of sensitive information to be detected.

Examples: Customer Phone Number, Resident Registration Number, Internal Project Name, etc.

Regular Expression Pattern

Define sensitive data patterns using regular expressions (Regex) to detect matching content.

When the system detects text matching a registered pattern, the details can be viewed in the Training Data Quality column.

Training Data Quality Management