Welcome to
Help Desk

Product Updates
Training
Support
Ideas Contact Support

AI-Driven Metadata Tags – Trainable Classifiers

Egnyte provides powerful capabilities to create and manage metadata tags. The metadata tags can then be used to search, sort, and filter files and folders. Trainable classifiers provide a new way to apply metadata tags at scale by combining the power of AI-based classification and Egnyte’s robust metadata tagging framework.

This feature allows Power Users and Administrators to supply seed files that are used by the classifier to learn what the target files contain. Once trained, the classifier can then apply a user-defined tag to all matching files within the folder(s) specified by the user. Any Power User that has Full permissions on a folder can create a new classifier and apply it to any folders that they have full permissions on.

Once a classifier has been created and applied to a folder (or a set of folders), all matching files or folders will get the tag automatically applied to them. In addition, any new files or folders, that are added in the future, will be processed by the classifier and the tags will be automatically applied to any items that match. Multiple classifiers can be created and applied to the same folder if needed.

A user can also see the list of all classifiers already created (including ones created by other users), and can apply any existing classifier to their folders (provided they have full permission on those folders).

This feature is available for Platform Enterprise plans.

Skip Ahead To...

Creating A Classifier

Train The Classifier

Continue Training A Classifier

Activating Or Deactivating A Classifier

Edit Or Delete A Classifier

Searching For Tagged Files

Limitations

 

Creating A Classifier

A Power User or an admin can create a classifier.

  1. Navigate to the Settings > Metadata Tags > Trainable Classifier Tags menu.
  2. Click on Add new Classifier.
    Trainable Classifiers 1.png
  3. On the Classifier details screen, user can specify the following:
    • Name: Name of this classifier.
    • Smart tag: It is the metadata tag that will be applied to matching files.
    • Description (Optional): Description for the trainable classifier.
    • Folder(s): These are the associated folders in which, files will be processed and tagged by the classifier. In order to associate folders with the classifier, click on the Select Folder(s) button and choose the folders.

      A user can only choose folders where they have full permissions or higher.

    • Seed content: It is a folder that contains all the files that will be used to train the classifier. In order to provide the seed content for training the classifier, click on the Select Folder button in the seed content section.

      - Minimum of 20 files are required in the selected folders for the classifier to be created.
      - In addition, at least 5 files must be provided as seed for training the classifier.
      - Egnyte recommends using at least 20 files for seed.

  4. Once the folders and seed have been specified, click on Save Changes.
    Trainable Classifiers 2.png

The screen would change to indicate that the classifier is being created. The classifier status would say Creating, and Estimated Accuracy would show Unknown. At this point, the classifier is processing the seed files.

Once the processing is complete, the status will change to Inactive. It may take several minutes depending on the length and complexity of files. The user may close this screen and come back later to check the status.

Trainable Classifiers 3.png

Back to Top...

Train The Classifier

Once the classifier status has changed to Inactive, the user can start interactively training the classifier to increase its accuracy.

  1. Click on the Start training link.
    Trainable Classifiers 4.png
    It will start a training session. Each training session consists of manually providing feedback on five files.
    Trainable Classifiers 5.png

    It is highly recommended that the user completes at least 4 training sessions (20 files) to get a better measure of classifier accuracy. The reported estimated accuracy of the classifier fluctuates widely for the first few sessions.

  2. After clicking on Start Training button, the classifier would show a file in the preview area and ask whether this file belongs to the class of documents the user is trying to train the classifier to recognize.
    • Click Yes or No as appropriate.
    • Users can also click on Stop Training to abandon this training session. They can start a new training session later as described in the following sections.
      Trainable Classifiers 6.png
  3. After responding to the Yes or No questions for five files, the user can choose to Complete the training session. Completing a session would save the changes and display an estimate of the classifier accuracy.
    The user can choose to continue training by clicking Next training session, or finish training, or Activate the classifier which automatically finishes training and starts tagging all files that match within the folders they had selected.
    Trainable Classifiers 7.png

Back to Top...

Continue Training A Classifier

The user can see the list of all AI trainable classifiers by navigating to:

  • Settings > Metadata Tags > Trainable Classifier Tags

To further train a classifier created earlier:

  • Click on the classifier's name to expand it, and then click on Continue Training next to the Estimated Accuracy.
    Trainable Classifiers 8.png

Continue training until the desired accuracy is reached. Egnyte recommends training every classifier for at least four sessions. Estimated accuracy measure may fluctuate widely during the first few training sessions.

Back to Top...

Activating Or Deactivating A Classifier

Activating a classifier will start applying tags to all files that match the classifier and are in the specified folders. An inactive classifier can be activated by clicking on the Activate link next to the status.

Trainable Classifiers 9.png

Once activated, the classifier will continue to examine any new files being added to those folders and apply tags as needed.

Trainable Classifiers 10.png

Similarly, an Active classifier can be deactivated by clicking on the Deactivate link.

Trainable Classifiers 11.png

Back to Top...

Edit Or Delete A Classifier

A classifier can be edited by the creator or by an admin.

To edit a classifier:

  • Navigate to Settings > Metadata Tags > Trainable Classifier Tags, and click on the Edit icon for the specific classifier to edit.
    Trainable Classifiers 12.png

Once the edit dialog opens, the user can modify the Name, Smart Tag, Description, and associated folders.

Changing the tag does not affect any of the files that are already labeled with the old tag. The new tag is applied only to the files tagged after the change is made.

If a folder is removed from the scope of this classifier, any files already tagged within that folder remain unaffected. However, no new files will be tagged on the folder that is now removed from the classifier scope.

Similarly, the user can also delete a classifier by clicking on the delete icon.

- Editing a classifier does not affect any files that have already been tagged. Any changes made only take affect for new files.
- If a classifier is deleted, any files that have already been tagged by the classifier retain their tags.

Back to Top...

Searching For Tagged Files

Once files are tagged, they can be searched using the Egnyte’s search functionality.

Additionally, if a file is opened in the preview pane, the user can also see & edit all tags assigned to the file in the metadata section.

Trainable Classifiers 13.png

Back to Top...

Limitations

  • The maximum number of seed files that will be considered is 100. If the selected folders contain more than 100 files, the additional files will be ignored.
  • Estimated accuracy measurement may fluctuate widely until at least four training sessions are complete.

Back to Top...

Was this article helpful?
0 out of 0 found this helpful

For technical assistance, please contact us.