Create and Manage a Case Classification Model

When you choose to generate case classification suggestions based on context recognition, you must configure a Coveo ML CC model to render classification suggestions in cases. Coveo ML CC models learn from support cases that have been correctly classified to provide classification suggestions on cases that haven’t yet been classified.

Example

On your support website, users can contact your support team by filling a support case. Under the hood, your Contact Support form contains a Case Classification feature that’s been configured to provide suggestions based on context recognition, which means that the feature is powered by a Coveo ML CC model.

A user accesses your support website to fill a support case and enters the following information:

  • Case Subject: Speedbit watch connection problem

  • Case Description: My Speedbit watch doesn’t connect with my smartphone.

Since the model has previously learned from the terms and classifications used by other users when filling their support cases, the model suggests the following classifications to the current user:

  • For the @producttype field: Speedbit, Watch

  • For the @issuetype field: Connection, Bluetooth

Prerequisites

Before creating a Coveo ML CC model, ensure that the support cases you want to use to train the model meet the following data requirements:

Closed Support Cases

To help the model train from quality data, we recommend that you use closed support cases when selecting the cases that must be used by the model. This allows the model to learn only from support cases that have already been resolved.

The Apply additional filters section of the model configuration allows you to segment the support cases on which the model bases its training.

Example

You want the Coveo ML CC model to base its training only on closed support cases.

Therefore, in the Apply additional filters section, you enter sfstatus is equal to Closed.

Sufficient Data in the Training Fields

The text fields of the support cases from which you want the model to base its training must contain at least 10 characters to be considered by the model.

Example

When configuring your Coveo ML CC model, you chose to train your model based on the subject and description fields.

Therefore, the model only takes into account the support cases that contain a minimum of 10 characters in both the subject and description fields for its learning process.

Sufficient Classified Support Cases

Each field value that is to be predicted by the model must be part of at least 500 indexed support cases.

Example

When configuring your Coveo ML CC model, in the Fields to predict section, you chose to provide predictions for the caseType field.

In the support cases from which you chose to train the model, the caseType field has three possible values: issue, request, and incident.

Among the available support cases:

  • 1007 have been classified as issue

  • 1885 have been classified as request

  • 283 have been classified as incident

Since there are only 283 cases that have been classified as incident, the model only provides issue and request as classification values for the caseType field.

English Support Cases

The value of the index field that identifies the language of the support cases must be English for the support cases to be considered by the model.

Create a CC Model

  1. On the Models (platform-eu | platform-au) page, in the upper-right corner, click Add model to open the Add a Machine Learning Model panel.

  2. Under Name, enter a meaningful display name for the model.

  3. Under Model type, select Case Classification, and then click Next.

  4. In the Learn from section, select the sources that contain the support cases from which the model should base its training.

  5. (Optional) In the Apply additional filters section, you can specify a condition to segment the cases that the model should learn from.

  6. (Optional) In the Edit default filters section, you can edit the default index fields that the model must use to identify the language and creation date of the support cases.

  7. Click Next.

  8. Select the index field corresponding to the unique identifier of your cases (e.g., sfcasenumber).

  9. Select the case fields that the model must use for its learning process (typically subject and description).

    Notes
    • For the model to learn from the fields chosen in this step, the scoped fields must contain between 10 and 3000 characters.

    • While we recommend using the subject and description fields, you can use any other field that contains information that tends to solve cases, such as resolution, summary, and problemDescription.

  10. Click Next.

  11. In the Specify the fields for which the model should provide predictions section, in the drop-down menu, select the fields that the model should provide predictions for (e.g., producttype and issuetype). Note that the selected fields must be of the Facet type, and can’t be of the Multi-Value Facet type.

    Tip
    Leading practice

    We don’t recommend selecting fields for which generic values are available (e.g., other or default).

  12. Click Next.

  13. In the Summary section, review your Coveo ML CC model configuration.

    Important

    Once the model has been created, you cannot modify its configuration. If necessary, however, you can modify its name.

  14. Click Add model.

Some of the support cases that have been targeted while configuring the model may later get filtered out because they can’t be used for training purposes. See CC Model Case Filtering for details.

CC Model Case Filtering

Some of the support cases that you scoped while configuring the model may get filtered out because they can’t be used for training purposes.

The model performs the following filtering process to ignore support cases for which:

  • Duplicated unique identifiers have been found. When the model finds cases with a duplicated unique identifier, it uses only one of them.

  • The selected fields to predict aren’t of the string data type.

  • The selected fields to predict have only one or no possible values.

  • The selected fields to predict contain empty values.

  • The selected training fields contain empty values.

  • The selected training fields aren’t of the string or list of strings data types.

  • The selected training fields contain less than 10, or more than 3,000 characters.

You can replicate how the model would filter some of the targeted support cases by using the Content Browser (platform-eu | platform-au) feature of the Coveo Administration Console.

Example

You plan on creating a CC model with the following configuration:

  • Support cases from the My Support Cases Source.

  • Uses support cases that have the closed status.

  • Uses support cases that have been created in October, November, and December of 2021.

  • Learns from support cases in English.

  • Learns from support cases the have the casenumber field as the unique identifier.

  • Uses the description and subject as training fields.

  • Uses the producttype and issuetype as fields to predict.

To see the indexed support cases that will be used by the model, you can enter the following query in the Content Browser:

(@status=="closed" @language=="English" @casenumber @sfcasenumber @subject @description @producttype @issuetype) ((@source=="My Support Cases Source") (@year==2021) (@month==(12,11,10)))

Delete a CC Model

  1. On the Models (platform-eu | platform-au) page, click the ML model that you want to delete, and then click More > Delete in the Action bar.

  2. In the Delete a Model panel that appears, click Delete model.

Review Active Model Information

On the Models (platform-eu | platform-au) page, click the desired model (must be Active), and then click Open in the Action bar (see Reviewing Coveo Machine Learning Model Information).

Reference

"Status" Column

On the Models (platform-eu | platform-au) page of the Administration Console, the Status column indicates the current state of your Coveo ML models.

When inspecting the Status column, a given model is either in the Active or Inactive state. Additional information can be displayed depending on the model’s current state.

status column

The following table lists the possible model statuses, their definitions, and their status colors as shown in the Administration Console:

Status Definition Status Color
Active The model is active and available.
Inactive The model isn't available.
Update in queue Waiting to process a scheduled update or configuration change.
Updating The model is being rebuilt based on a new configuration.
Waiting The model is in the building queue.
Building The model is currently being processed.
Degraded The model is active, but has some limitations. Additional information is available in the Error section of a model (see Review Coveo Machine Learning Model Information).
Failed The model couldn't be built with the requested configuration. Additional information is available in the Error section of a model (see Review Coveo Machine Learning Model Information).
Update failed The model couldn't be updated with the requested configuration.
Error An error prevented the model from being built successfully.

Filtering Options

When configuring a Coveo ML CC model, you can specify additional filters to adapt how the model should learn from the data it receives.

The Apply additional filters section allows you to specify a condition to segment the cases that the model should learn from.

The Edit default filters section lets you modify the default filters used by the model to identify the language and creation date of the support cases.

"Apply Additional Filters" Section

In this section, you can optionally specify a condition to segment the cases that the model should learn from.

To avoid the model building on inaccurate information, we recommend that you scope only support cases that have been closed.

Example

You want the Coveo ML CC model to base its training only on closed support cases.

Therefore, in the Apply additional filters section, you enter sfstatus is equal to Closed.

To specify an additional filter

  1. In the Apply additional filters section, click Add-Filter.

  2. In the Add a Condition panel that opens, in the Field name input, enter the name of the field that you want to use to segment cases.

  3. In the Select an operator drop-down menu, select the desired operator.

  4. In the Value input, enter the value of the field on which you want to segment the cases.

  5. Click Add filter.

"Edit Default Filters" Section

In this section, you can optionally edit the default index fields that the model must use to identify the language and creation date of the support cases.

By default, the model uses the following field values:

  • Language: The value of the language field.

  • Date: The value of the date field.

However, your support cases may use the value of different fields to identify their language and creation date.

Example

The support cases you want to use to train your Coveo ML CC model use the following field values to identify their language and creation date:

  • Language: The value of the sflanguage field.

  • Date: The value of the sfcreationdate field.

Therefore, you specify these fields in the Edit default filters section to instruct the model to use the sflanguage and sfcreationdate values rather than the default ones.

To edit the default language and date filters

  1. In the Language drop-down menu, select the field that must identify the language of the support cases.

    Important

    The value of the field that identifies the language of the support cases must be English for the support cases to be considered by the model.

  2. In the Date range section:

    • In the first drop-down menu, select the field that must identify the creation date of the support cases.

    • In the second drop-down menu, select the period of time for which the model should learn from support cases. By default, the model learns from the support cases created in the last 6 months. However, you can select one of the available preset ranges, or enter a custom date range.

    Tip
    Leading practice

    We recommend that you use a dynamic date range (i.e., Last 3 months, Last 6 months, or Last year) to train your model to ensure that it learns from the latest available data.

    Using a fixed date range (e.g., from 2022/01/01 to 2022/06/01) results in the model always retraining on the same data.

Required Privileges

By default, members of the Administrators built-in group can manage Coveo ML CC models.

The following table indicates the privileges required to manage Coveo CC models (see Manage Privileges and Privilege Reference).

Action Service - Domain Required access level

View models

Machine Learning - Models

Search - Query pipelines

View

Manage models

At least one of the following privileges:

  • Machine Learning - Allow content preview

  • Search - View all content

Enable / Allowed

Machine Learning - Models

Edit

Search - Query pipelines

View

What's next for me?