Add-on module required
You will only have access to the features described in this article if you have purchased the Classification Intelligence add-on module with your Records365 subscription.
To Train a model you need to be assigned to either the Application Administrator or Records Manager role in Records365.
The training process leverages records in Records365 that already have a business classification to build a model that can, with high confidence, predict a record category (business classification) for new content discovered by Records365.
Currently content in Records365 can be assigned a record category (business classification) in the following ways:
- Metadata based, through the use of the automated Rules Engine
- Manual, through the manual assignment of record categories and the Reschedule feature
- Content based, through the use of the Classification Intelligence add-on
Once a category is added to the machine learning model successfully, it will show in the Training page as Included. For every subsequent retraining, that record category class will be automatically included unless the user removes the category before training at which point the record category will then be "Queued for Exclusion" and then removed once the user has triggered a model retrain.
Some areas of a file plan may be better suited to machine learning, and so two or more record categories can be targeted for inclusion in a model through the training process.
For a successful training, the following criteria must be met:
- At least two record categories must be selected.
- A record category is eligible for inclusion into a model if it has at least 50 records assigned.
- At least 10 documents are of a file type that can be analyzed.
- At most, the newest 500 documents discovered by Records365 will be used for training.
To build a model with a selected set of record categories, perform the following:
- In the left-hand navigation menu, click on the Intelligence section and select Training.
- Navigate to the Training tab. In there, you will see any record categories in your file plan that Records365 is currently trained on.
- Expand your file plan until the desired record categories are shown, or search for a particular record category in the search field.
- To add a record category, click the box next to the category and then click the Include Categories button (referred to as Disposal Classes in some regions) to add the record category to be Queued for inclusion.
- Once all of the record categories have been added, click the Start Training button to begin training a new Classification Intelligence model with the changes to categories to be included and/or excluded.
- To remove a record category
- To remove a record category, click the box next to the category
- Click Exclude Categories to add the record category to be Queued for Exclusion.
- Once all changes to which record categories are queued for inclusion and exclusion, click Start Training to retrain with those changes applied, thereby removing the category.
- Note: If you remove record categories that are part of your current model, Records365 will no longer be able to make suggestions for the removed record categories.
Note: Only those record categories set to be Available for Classification Intelligence will be available for selection. For more information on managing record categories see the documentation on the File Plan page.
You will not be able to train Records365 in the following scenarios:
- an insufficient number of record categories have been selected
- a training run is currently in progress
- a training run has already been performed in the last 24hrs
All model training is performed in the English language. If documents are of another language, there may be a reduction in prediction health. Non-english documents are not excluded from training.
Upon successful training, you will be able to view your trained categories as well as their classification skill. With this information, you can assess categories suitable for automatically applying the suggested categories. For more information on enabling auto-apply on record categories please see the Auto-apply page.
The Trained Models tab is where you can view the results of all training runs that have been carried out in Records365.
You can gain the following insights about each model that Records365 is trained on:
- Which model is active, and is currently being used to offer suggestions to records
- The user who initiated the training run
- When the resultant model was created
- How healthy the resultant model is
- The status of the model, the latest successful model will be promoted to have an Active status.
To view the details for a model, like who created it, the accuracy and even the record categories that it includes;
- Click the Experiment ID for a model from the Trained Models tab
- For those record categories covered by the model, see the Trained Categories tab.
The training set is the set of records that were used to represent a record category when Records365 is being trained.
- In the left-hand navigation menu, click on the Intelligence section and select Training
- Click on Training Set
Use this page to review the records that were used to train Records365. You may find that a record category that is performing poorly was trained using documents that are not representative of that category of record.
The training set will include the following records
- a maximum of 500 records that are assigned a record category that is included in the model
- where more than 500 records are assigned a record category, only the 500 records most recently modified in the content records will be used
- only those records assigned a record category that is part of the model will be included
- records assigned the record category using all methods except Classification Intelligence Auto Apply
- records that have content that are of a type that can be analyzed by Classification Intelligence
- records where the Classification Intelligence suggestion was rejected, the record category assigned at time of rejection will be displayed in the Training Set page
For subsequent training runs the set of documents included in the training continues to grow such that
- records where the Classification Intelligence suggestion was rejected will continue to be added
- records created since the last training run that are assigned a record category from the new model will continue to be added, up to the point where the maximum number of records per category is reached
A training set will be reset and recreated in the following scenarios
- it has been 14 days or more since the previous training
- a training is performed which contains a different set of record categories to the previous training
- where the training set contains > 40% of records that have changed since the last model was created
Only the training set for the model that is currently active is available for review.