Reusing saved models

Saved models in Review Center provide the ability to take the knowledge, or the training, from one Review Center queue and re-use it in another queue. A saved model contains how many times a word occurred and what those words are, essentially remembering what was relevant, what was irrelevant, and how those were defined. With that information, you can use it to find relevant documents in a new queue or workspace.

See these related pages:

What a saved model contains

A saved model contains the training from a previous Review Center queue, which includes:

What words occurred.
How many times words occurred.
Which words generally predicted relevance or irrelevance, and a “relevance score” for each word.

The model does not contain any actual documents from the original queue. Instead, it contains relevance predictors in an encrypted, digested form that cannot be accessed by human users.

How predictions work with multiple models

When a saved model is linked to a queue, Review Center makes its predictions by averaging the relevance scores stored in the linked model and the local model. The local model contains the scores for words based on all coding decisions within the queue.

If there are several linked models, Review Center first takes the mean of the scores within the linked models, then averages that result against the local model's scores.

For example, if linked model 1 assigns a relevance score of 30% to the word "housing," and linked model 2 assigns it a score of 80%, this averages to 55%. If you start a brand new queue with both of these models attached, Review Center scores the word "housing" at 55%.

If you attach those same models to a queue that already contains some coding decisions, the local model may already have an entry for the word "housing." If the local model scores it at 50%, Review Center averages 50% with 55%. This gives the word "housing" a final score of 52.5%.

Privacy considerations when reusing saved models

Saved models can be shared across workspaces within the same instance, regardless of client domain, as long as the person sharing the model has access to both workspaces.

There is some risk that the model will reveal aspects of its training indirectly based on how it classifies. For example, if a document that it predicts as relevant contains the name "Jennifer" and nothing else, users can assume that the original queue had "Jennifer" in the source data and that it was considered relevant. However, the source documents themselves are not actually revealed. The model does not store any identifying information such as the name of the original workspace, the name of the queue, or the control numbers of the documents that trained the original model.

For more information about permissions related to saved models, see Review Center security permissions.

Common use cases for saved models

If you handle cases with similar document types or subject matter, saved models can help you jump-start a new case and start reviewing relevant documents more quickly. Instead of training a new model from scratch, you can link one or more saved models to a new queue and immediately start coding documents that the model predicts as relevant. After the new queue is underway, you can choose either to continue with both the saved models and the local model built from newly coded documents, or you can remove the links to the saved models and continue coding with only the local model.

Some use cases include:

Anti-trust cases in different jurisdictions.
Cases with similar subject matter such as medical-based reviews, toxic workplace, or bribery.
Serial litigations.
Internal investigations for improper or risky behavior.

You can also use saved models to find particular document types across cases. Examples of these include:

Culling junk documents such as spam emails.
Culling office noise such as vacation chatter and party invitations.
Finding specific categories of documents such as contracts or financial paperwork.

Creating a saved model

You can save a Review Center model that has at least five positive and five negative documents coded. Queues with more documents coded will have more fully developed models, so we recommend saving models from late-stage or completed Review Center queues.

For a list of required permissions, see Review Center security permissions.

To create a saved model from an existing Review Center queue:

From the Review Center dashboard, select the queue.
On the right side of the Queue Summary section, click the three-dot menu icon ().
Select Save as New Model.
Fill out the following fields:
1. Name—the name of your saved model. If you plan to save the model from this queue multiple times, consider including a version number or a date.
2. Description (Optional)—identifying features such as the model's purpose or what workspace and queue it originated from.
Click Save.

When the save completes, a green success banner appears at the top of the dashboard.

Note: The Saved Models feature was released in March 2025. If you want to save a model from a queue older than that, refresh the queue first.

Creating from queues with linked models

If you create a saved model from a queue that already has a saved model linked to it, the newly saved model will contain the training from both the linked model and the local model.

Linking a saved model to a Review Center queue

After creating a saved model, you can link it to another Review Center queue to jump-start the coding predictions for the new queue. For most situations, we recommend linking models to a newly created or early-stage queue. However, it is possible to link them at any stage. Any documents that are already coded within the destination project will add to the relevance predictions, but they are not required for the model to build.

To link a saved model when creating a new queue, see Creating a new queue from a template.

To link a saved model to an existing queue or to switch models:

From the Review Center dashboard, select the queue.
On the right side of the Queue Summary section, click the three-dot menu icon ().
Select Edit.
Next to Saved Model, click Select.
The model selection options appear.
Select the model or models you want to link.
Click Apply.
Click Save.
Refresh the queue to make the changes take effect.

Removing a linked model from a queue

After you have coded enough documents in the queue for the local model to build, you can remove the linked model at any time. After you remove it, the queue's relevance predictions will be calculated only from coding decisions within the queue.

To remove a linked model from a queue:

From the Review Center dashboard, select the queue.
On the right side of the Queue Summary section, click the three-dot menu icon ().
Select Edit.
Next to Saved Model, click Clear.
Click Save.
Refresh the queue to make the changes take effect.

How linked models behave with ARM

When using the Archive, Move, Restore (ARM) tool, linked models will be retained if you archive and restore within the same instance. However, if you attempt to restore a workspace with a linked model in another instance, you will receive an error and will not be able to start or refresh the queue. To start or refresh the queue, remove the linked model from it.

Managing saved models

You can access your saved models on the Saved Models tab in your workspace. This tab shows all models that originate in this workspace, as well as all models that have been copied to this workspace from outside.

The tab shows the following for each model:

Name—the name of the saved model.
Description—any description given to the model when it was saved.
Created Date/Time—the date and time the model was originally saved or copied to this workspace.
GUID—the model's Globally Unique Identifier (GUID). This ID stays the same for a model regardless of which workspace it appears in or whether it has been renamed.

If you have several saved models to manage, you may want to create a workspace to serve as a central model library. Copying all saved models to this library workspace, then re-sharing them to individual workspaces as needed, allows you to manage all of your saved models in one place.

After a model has been copied to the library workspace, it can be safely deleted from its original workspace. Copies of models are not linked to the original copy.

Copying saved models to another workspace

Models can be copied to workspaces within the same instance, regardless of client domain. To copy a model, you must have access to the destination workspace.

For more information on permissions, see Review Center security permissions. For more information on model sharing and privacy concerns, see Privacy considerations when reusing saved models.

To copy a model and make it available for use in another workspace:

On the Saved Models tab, click the Copy To icon ().
A workspace list appears. This list includes any workspace you have access to in the instance.
Select the workspace or workspaces you want to share to, then click Copy To.

After copying, each workspace has its own separate copy of the model.

Editing or deleting saved models

Every saved model is independent. Editing or deleting a saved model in one workspace does not affect shared copies in other workspaces, nor does it affect the Review Center queue the model was saved from.

To edit a saved model:

From the Saved Models tab, click the Edit icon () beside the model.
Edit the fields you want to change.
Click Save.

To delete a saved model:

On the Saved Models tab, click the Delete icon () beside the model.

A confirmation message appears.
Click Delete.

Saved models from Active Learning

If your workspace contains trained models from the older Active Learning application, these have been automatically converted to saved models for Review Center. You can view these on the Saved Models tab alongside the other models.