How to retrain a classifier in content explorer

  • Straipsnis

A trainable classifier is a tool you can train to recognize various types of content by giving it samples to look at. Once trained, you can use it to identify items for application of Office sensitivity labels, communications compliance policies, Microsoft Purview data loss prevention policies (DLP) and retention label policies.

This article shows you how to improve the performance of custom trainable classifiers by providing them with more feedback.

To learn more about the different types of classifiers, see Learn about trainable classifiers.

Note

Pre-trained classifiers cannot be retrained.

Tip

If you're not an E5 customer, use the 90-day Microsoft Purview solutions trial to explore how additional Purview capabilities can help your organization manage data security and compliance needs. Start now at the Microsoft Purview compliance portal trials hub. Learn details about signing up and trial terms.

Permissions

To access classifiers in the Microsoft Purview compliance portal:

  • the Compliance admin role or Compliance Data Administrator is required to train a classifier

To use classifiers in the following scenarios, you will need the permissions listed:

Scenario Required Role Permissions
Retention label policy Record Management
Retention Management
Sensitivity label policy Security Administrator
Compliance Administrator
Compliance Data Administrator
Communication compliance policy Insider Risk Management Administrator
Supervisory Review Administrator

Overall workflow

Important

You provide feedback in content explorer to automatically apply retention label policies to Exchange items and that uses the classifier as a condition. If you don't have a retention policy that automatically applies a retention label to Exchange items and that uses a classifier as a condition, stop here.

As you use your classifiers, you may want to increase the precision of the classifications that they're making. You do this by evaluating the quality of the classifications made for items it has identified as being a match or not a match. After you make 30 evaluations for a classifier, it takes that feedback, and automatically retrains itself.

To understand more about the overall workflow of retraining a classifier, see Process flow for retraining a classifier.

Note

A classifier must already be published and in use before it can be retrained.

How to retrain a classifier in content explorer

Select the appropriate tab for the portal you're using. To learn more about the Microsoft Purview portal, see Microsoft Purview portal. To learn more about the Compliance portal, see Microsoft Purview compliance portal.

  1. Sign in to the Microsoft Purview portal > Information Protection > Explorers > Data explorer.

  2. Under the Choose a classifier or a label list, expand Trainable classifiers.

Important

It can take up to eight days for aggregated items to appear under the trainable classifiers heading.

  1. Choose the trainable classifier you want to give feedback on.

  2. In the All locations list, open a folder that shows that matches have been found.

  3. Choose an item and open it.

    Tip

    You can provide feedback on multiple items simultaneously by choosing them all and then choosing Improve classification in the command bar.

    Note

    If an item has an entry in the Retention label column, it means that the item was classified as a match. If an item doesn't have an entry in the Retention label column, it means it was classified as a close match. You can improve the classifier precision the most by providing feedback on close match items.

  4. Choose Provide feedback.

  5. In the Detailed feedback pane, if the item is a true positive, choose, Match. If the item is a false positive, that is, it was incorrectly included in the category, choose Not a match.

  6. If there's another classifier that would be more appropriate for the item, you can choose it from the Suggest other trainable classifiers list. This will trigger the other classifier to evaluate the item.

  7. Choose Send feedback to send your evaluation of the match, not a match classifications and suggest other trainable classifiers. When you've provided 30 instances of feedback to a classifier, it will automatically retrain. Retraining can take from one to four hours. Classifiers can only be retrained twice per day.

    Important

    This information goes to the classifier in your tenant, it does not go back to Microsoft.

  8. Open Trainable classifiers.

  9. The classifier that was used in your Communications compliance policy will appear under the Re-training heading.

  10. Once retraining completes, choose the classifier to open the retraining overview.

    classifier retraining results overview.

  11. Review the recommended action, and the prediction comparisons of the retrained and currently published versions of the classifier.

  12. If you satisfied with the results of the retraining, choose Re-publish.

  13. If you aren't satisfied with the results of the retraining, you can choose to provide more feedback to the classifier in the Content Explorer interface and start another retraining cycle or do nothing in which case the currently published version of the classifier will continue to be used.

Details on republishing recommendations

Here's a little information on how we formulate the recommendation to republish a retrained classifier or suggest further retraining. This requires a little deeper understanding of how trainable classifiers work.

After a retrain, we evaluate the classifier's performance on both the items with feedback and any items originally used to train the classifier.

  • For built-in models, items used to train the classifier are the items used by Microsoft to build the model.
  • For custom models, items used in the original training the classifier are from the sites you had added for test and review.

We compare the performance numbers on both sets of items for the retrained and published classifier to provide a recommendation on whether there was improvement to republish.

See also