Page tree
Skip to end of metadata
Go to start of metadata

There is an administrative user interface for configuring the relevancy models in your system. In order to use it, your project must include the module called "relevancy." If it is installed, you can view the Relevancy Administration UI by going to the Attivio Business Center UI and clicking the Manage Relevancy button. (You can also access the Relevancy Administration UI directly from the Admin UI's navigation bar by clicking the Relevancy Model Administration link.)

Working with Relevancy Models

When you first visit the Relevancy UI, you are brought to the Relevancy Models page. This page shows a list of all of the models defined in your Attivio system. Here you can create new models as well as modify or delete the existing ones. While working in the Relevancy UI, you can return to the Relevancy Models page by clicking the View Relevancy Models button at the top of any of the pages ().

The list of models is in the main part of the page. When you select a model, the inspector on the right side of the page shows the details about that model, including its name, its type, and any additional parameters. The information displayed is:

IconAn icon for the model appears in the top-left corner of the inspector. If the model has been published, a yellow dot appears in the top-left corner of the icon to indicate so. If it is statically defined, a lock icon appears in the icon's bottom-left corner.
NameThis is the name of the model.
Status IconAn icon appears to the right of the model's name if there is an error () or if it is in the process of being trained ().
Edit ButtonThe Edit button, an icon in the shape of a pencil (), lets you edit the properties of the model.
Inspector MenuA pop-up menu with additional commands to use on the model is available under the gear () icon. The commands available here depend on the properties of the model — only commands applicable to the model being displayed are shown.
Model TypeThis shows whether the model is a user-defined or ML one.
Accuracy ScoreFor ML models, this is the percentage score generated by the machine-learning algorithms expressing how accurately the model's rankings match those provided by the signal data.
Latest VersionThis is the version number of the latest version of the model. Note that this version may not yet have been published.
Time Last ModifiedThis is the time that the model was last modified, in other words, the modification date for the latest version.
Published VersionThis is the version number of the published (active) version of the model, if it has been published.
Time PublishedThis is the time the model was last published, if any.
Error TypeIf the model is in an error state, this shows the type of error that occurred.
Error MessageIf the model is in an error state, this shows a description of the error that occurred.
History TabThe History tab of the inspector shows all versions of the model, along with the status, accuracy score, modification date, and publication state of each. You can view the features and weights that make up the model, publish a particular version, or delete individual versions using the buttons on the right side of the table.
Learning TabFor ML models, the Learning tab shows the parameters used when training the model: when to automatically publish it, how many versions to keep, which signal types should be used when training it, the maximum age of the signals, and which features to include.

You can filter the list of models to find the ones you're interested in by using the search box to enter part of the model's name in the search box; click the magnifying glass icon () or press the enter key to perform the search.

Creating Relevancy Models

See Configuring Relevancy Models for manually configuring relevancy models.

See Training Relevancy Models for automatically training relevancy models.

Editing Relevancy Models

To edit a relevancy model, select it in the list so that it appears in the inspector on the right-hand side of the window.


You can use the model name search bar to help locate the model you want to edit.

You can then click the Edit () button at the top of the inspector to open the Edit Relevancy Model wizard where you can modify its values, similarly to when creating new models, except that the name of a model cannot be edited. When you are finished editing the model, a new version of the model is created. If the model you edited is a machine-learning model, training of the new version is begun immediately.

Copying Relevancy Models

To copy an existing relevancy model, select it in the list so that it appears in the inspector on the right-hand side of the window. Then choose the Copy command from the inspector's menu (the gear icon — ). The Copy Relevancy Model dialog appears in which you can enter a name for the new model. Click the Copy button to complete the process. The new model will appear in the list.

Publishing Relevancy Models

To make a relevancy model available for use by queries, it must be published. At any given time, only one version of a particular model can be published — publishing one version will effectively unpublish the previously published one.

To publish a particular version of a relevancy model, select the model in the list so that it appears in the inspector on the right-hand side of the window. Find the version you want to publish in the History tab at the bottom on the inspector and click the Publish button (the yellow dot — ) to the the right of of it to activate it.

If you have a machine-learning model that is configured to be automatically published, if it is trained and the accuracy score meets the specified threshold it will be published without any intervention on the part of the administrator. If the score doesn't meet the threshold, the administrator can still choose to publish it manually as described here.

Retraining Relevancy Models

When a machine-learning model is created or edited, it is automatically submitted to the machine-learning algorithms to be trained. However, you can explicitly train the model at any time. You might want to do this, for example, if the signal data or the contents of your index have changed significantly. To do so, select the model in the list so that it appears in the inspector on the right-hand side of the window. Choose the Retrain command from the inspector's menu (the gear icon — ) to initiate the retraining. The status of the model will change in the inspector to show that it is being trained and a progress icon () will appear next to its name. When the training has completed, the status will be updated to show that it has been trained. If there was an error, an icon () appears next to the model's name details about the error are shown in the inspector, below.

Deleting Relevancy Models

When deleting relevancy models, you have two options: you can delete specific versions of the model individually or you can delete the entire model all at once. In either case, select the model in the list so that it appears in the inspector on the right-hand side of the window.


Deleting models, whether an individual version or the entire model, is permanent and cannot be undone.

Deleting Individual Model Versions

You can delete an individual version of a model by clicking the trash-can icon () next to it in the History tab at the bottom on the inspector. (If a version is statically defined, then there will be no delete button. This is the case for the currently published version as well, unless it is the only version left, in which case it can be deleted.) A dialog box will appear to confirm the deletion.

Deleting All Versions of a Model

To delete all versions of the model, including the published version, choose the Delete All Versions command from the inspector's menu (the gear icon — ). Confirm the deletion in the resulting dialog box and the model will be removed entirely from the system. Any queries using this model will behave as if it wasn't specified in the query. Any signal data tagged with this model's name will continue to exist.



You cannot delete statically defined models (i.e., those defined in your project's configuration XML files). If you delete all versions of a model that was originally statically defined, only the original version will remain.




  • No labels