Thank you for your feedback.
Form temporarily unavailable. Please try again or contact docfeedback@servicenow.com to submit your comments.

Create and train a clustering solution

Log in to subscribe to topics and get notified when content changes.

Create and train a clustering solution

Group similar records into clusters so you can address them collectively or identify patterns.

Before you begin

  • Create or reuse a word corpus that is relevant to your solution.
  • Role required: admin or ml_admin

About this task

Predictive Intelligence supports training solutions in which the source data is protected by these types of encryption.
  • FDE (Full Disc Encryption).
  • Platform Encryption. When using Platform Encryption, ensure that the sharedservice.worker user has the same encryption context role that has been used for encryption.
Predictive Intelligence doesn't support training solutions in which the source data is encrypted by Edge Encryption.
Note: Clustering only supports English language processing.

In this example procedure, you are grouping similar incidents that have occurred recently to identify a major incident.

Procedure

  1. Navigate to Predictive Intelligence > Clustering > Solution Definitions.
  2. On the ML Solution Definitions list, click New.
  3. On the form, fill in the fields.
    Field Value
    Solution Template Select the Clustering Template.
    Label Enter a unique name for your clustering solution. For example, in this use case you could enter Group Incidents to a Major Incident.
    Name As you enter your solution Label value, this field automatically populates with a system-assigned name that is similar to your label value.
    Word Corpus

    Select an existing word corpus that is relevant to your solution. For example, in this use case you select a word corpus that has a title such as Incidents in the last 3 months.

    If you don’t have a relevant word corpus, follow the steps in Create a word corpus. When the word corpus is complete, you can select it from the Word Corpus field in your ML Solution Definition form.

    Note: The number of records per table for word corpus creation used in clustering solutions is limited to 300,000.
    Table

    Select the table that contains record types that you want to group into one or more clusters. For example, in this use case you select the Incident [incident] table as it contains incident records you want to group together for a major incident analysis.

    When you assign a table value, a link appears in the form that shows the number of records that match your current conditions.

    Fields Select one or more field types that help the system identify the records you want to include in your cluster. In this use case, we use Short description.
    Use Group By Select this check box if you want to group input records by a field before creating clusters.
    Note: Selecting this check box activates the Group By choice list below. If you don't select the check box, all table records are grouped into clusters.
    Group By

    Select a value from this list. When you do so, the system groups records into one or more clusters based on your selection. For example, in this use case you select Category. If your Incident [incident] table has 10 category types, the system groups each type into an individual cluster, rendering 10 clusters.

    Filters Add filter conditions to the Cluster Input Fields records you want included in your clusters.
    Note: The number of records for clustering is limited to 100,000.
    Refresh Frequency
    Select how often you want the system to group new and updated records into clusters.
    Note: The system pulls records based on the Group By filter conditions that you set on your clustering solution, if any.

    For example, if you select Every 15 minutes, the system identifies which records have arrived within that time frame and tries to assign them to the existing clusters, or creates a new cluster if possible. So let's say that 20 new records arrive. If 16 of the 20 records make it into an existing cluster and 4 don't, the system forms a new cluster for the 4 unassigned records.

    Recluster Frequency Select how often you want the system to discard all previous cluster results and recreate clusters from the beginning.
    Target Solution Coverage Use this 0-100 percentile field to filter out records that are less similar to each other. For example, set the target coverage value to 75 so your clusters include only up to the top 75% of similar records. If you're unsure which value to enter, start with 50. You can always change the values, retrain your solution, and review the results.
    Minimum number of records per cluster Enter the minimum number of records you want to allow in any cluster. The value you enter must be greater than or equal to 2.
  4. Click the appropriate button for your solution definition.
    OptionDescription
    Save Save your solution definition record so you can return to it later.
    Submit & Train Create your solution definition record and train it.
  5. If you submitted the solution for training, click OK on the Training Activation window to confirm.

Result

  • The system trains the solution and notifies you in real time when the training completes.
  • A scatter plot chart populates the Cluster Visualization tab of your ML Solution form, showing the top 50 clusters and their individual records. When you point to a cluster you can see its label, size, and quality percentile value. The cluster labels are ordered based on cluster quality. You can filter the results by using the two slide bars for cluster size and cluster quality, respectively.
Shows an image of the Clustering scatter plot chart, with the top 50 clusters and their individual records. When you point to a cluster you can see its label, size, and quality percentile value.

What to do next

Review the solution output on the Solution Statistics tab of your ML Solution. If you aren't satisfied with your clustering solution results, reconfigure the values you've set to your solution and retrain it until the results are to your satisfaction.
Feedback