Thank you for your feedback.
Form temporarily unavailable. Please try again or contact docfeedback@servicenow.com to submit your comments.
Versions
  • London
  • Kingston
  • Jakarta
  • Istanbul
  • Helsinki
  • Geneva
  • Store
Close

Create and train a solution definition

Create and train a solution definition

Specify the records used to train a predictive model, what fields trigger a prediction, and how often to retrain a solution.

Before you begin

Role required: admin or ml_admin

About this task

A predictive model is only as good as the data you use to train it. To select appropriate training records, familiarize yourself with the table database dictionary as well as the current quality of the record values to be used.

You must create a separate solution definition record for each predictive model you want to support. To expedite this process, you can copy a solution definition record and its configuration into a new form by clicking Copy Solution Definition from the context menu of the original record. You can use the new record to make further solution updates without reconfiguring the entire solution definition to change something minor.

Note:
  • If your data is encrypted by FDE (Full Disc Encryption), you can train solutions.
  • If the field you are using for training is encrypted using Platform Encryption, ensure that the sharedservice.worker user has the correct encryption context role that has been used for encryption.
  • Agent Intelligence does not support training if you are using Edge Encryption.

Procedure

  1. Navigate to Agent Intelligence > Solution Definitions.
    The system shows the current list of solution definitions.
  2. Click New.
    The system shows a blank solution definition form.
  3. Configure these field values per the guidance below.
    Field Value
    Solution template
    Select a template to fill in the form with preset values. For example, the Incident Assignment Template selects the Incident table and creates a filter that selects resolved or closed incidents within the last 12 months.
    Note: This field cannot have a value of None.
    By default, Agent Intelligence offers these solution templates.
    • Assignment Template
    • Category Template
    • Incident Assignment Template
    • Incident Category Template
    • Incident Priority Template
    • Priority Template
    Table Select the table containing the target records to be predicted.
    Domain On instances where domain separation is active, select the domain whose target records you want to predict. Create a separate solution definition record for each domain whose field values you want to predict.
    Label Enter a name for the solution record.
    Name The system generates the value of this read-only field based on the Label.
    Active Select whether the system uses the solution definition record to train solutions. You can only train active solution definition records.
    Filter
    Select the conditions you want to apply to the training records. To train a solution, the filter must return at least one record. If your filter returns no records, update it until it returns records for training.
    Note: The recommended number of records for training a good solution is from 30,000 through 300,000. If you submit more than 300,000 records, the most recent 300,000 records are used to train the solution. Use only authentic records from the database.

    The system provides a default filter when you select a Solution Template.

    A solution is only as good as the data you use to train it. In general, a good filter has these characteristics.
    • The training records are inactive and have task states that represent completing work within your standard process such as resolved or closed.
    • The training records only contain correct values for the target field. Filter out records with unreliable target field values. For example, if you are predicting the assignment group/category and you have assignment groups/categories in your historic incident data that are no longer used, add a filter to remove such incident records from the training.
    • The training records contain multiple examples of each target field value you want the solution to predict.
    • The training records include common variations of the input fields.

    For example, the Incident Category Template creates a filter with these conditions.

    [Created][on][Last 12 months] AND [Active][is][false] AND [State][is one of][Resolved | Closed]

    Do not use hard-coded dates as filters, as these filters are not updated when the solutions are retrained unless you update them manually before every retraining. Instead, use relative date filters, such as last 3 months, last 6 months, and last 12 months.

    Processing Language Select the dominant language of the dataset you are training on the solution definition. If the dataset language is English only, you can choose None, as English processing is applied to all datasets by default. For example, if you select Dutch, data is processed in both English and Dutch.
    Note: The term “processing” indicates some of the language-specific steps used as part of training a solution. For example, tokenizing words, removing stop words, and stemming.
    Input fields

    Select the input fields you want the solution to use to generate a prediction. The system provides default input fields when you select a Solution Template.

    In general, good input fields have these characteristics.
    • The fields are available to users when creating records. Configure the form to show all input fields.
    • The field data type is string. The more information a field provides, the more often a solution can make a prediction, and the more often predictions are accurate.
    • The field has a default value. The field should not have a blank value.

    For example, all default solution definitions use the Short description field as the input field.

    Output field

    Select the field whose value you want the predictive model to set. The system provides a default output field when you select a Solution Template.

    In general, a good output field has these characteristics.
    • The field is a choice field or a string field with a finite set of possible values.
    • The field has some casual connection to the input fields.

    For example, the Incident Category Template selects the Category field.

    Training frequency
    Select how often the system regenerates the solution based on records matching the Filter. Options include:
    • Run Once
    • Every 30 days
    • Every 60 days
    • Every 90 days
    • Every 120 days
    • Every 180 days

    By default, the system runs training once. This practice provides you time to review and update the solution definition as needed until it provides acceptable coverage and precision values.

    When your solution definition is fairly stable, consider scheduled trainings, as data can age over time, thus degrading the accuracy of your prediction model.

  4. Click the appropriate button for the solution definition.
    OptionDescription
    Activate & Train Activate and train a default solution definition.
    Submit & Train Create a solution definition record and train it.
    The Training Activation confirmation window appears.
  5. Click OK to confirm.
    The system schedules the solution for training with the nearest training service. When training is complete, the system uploads the solution as an Attachment record.

What to do next

Review the trained solution precision and coverage statistics.