Model Retraining

Triage is a machine learning model that requires continuous learning and improvement. To continue enhancing its performance, it's important for Service Desk Agents to reroute any incorrectly predicted tickets. When Triage makes incorrect predictions and tickets are rerouted by agents, this feedback is automatically collected and used for future retraining. Retraining the model with new data helps it "relearn" and improve its accuracy for future predictions.

What can you do with the Model Retraining module?

In this module, you can define the filtering criteria for tickets on which you would like to retrain the model.

For eg. Once the Triage model is deployed and you observe that Triage predictions are not accurate for certain cases where assignment group is predicted as A, B, or C. If you would like to retrain all those tickets where assignment group value is say A, B, or C, and contact type is say, Bot and Email. Then in this module, you can define these filtering conditions, so that Moveworks can retrain the model once every month automatically and improve the overall accuracy of the model.

Prerequisites

1. Before proceeding for Model retraining configurations, it is important to make sure that Triage model is actually trained and deployed on your org's production instance.
2. You have a good understanding of predicted values where prediction accuracy is low.

Configuring Model Retraining

There are two sections within this module:

1. Model Details: In this section, you need to input key details around your model such as Integration ID, model name, ticket types, additional features, etc.
2. Ticket Filters: In this section, you can define the ticketing filters for retraining.

Model Settings

1. Input the Integration ID. This is the integration id for the ITSM system. For eg. If Triage is running on Service Now ticketing, choose the integration ID corresponding to the Service Now.
2. Input the Model Name. This is the predicted field name. This should be the predicted field name, the model is updating, say assignment_group. Concatenated models should use + as the delimiter. ie. category+subcategory. Refer predicted fields from below:
   1. Assignment group = assignment_group
   2. Category = category
   3. Sub Category = subcategory
   4. Business service = business_service
   5. Service Offering = service_offering
   6. Configuration item = configuration_item
   7. HR Service = hr_service
   8. Components = components
   9. Call Type = call_type
3. Select the Ticket Type. This correspond to the type of tickets we’re intercepting for prediction. E.g. Incident, Call, Default.
4. Select Additional features, if applicable. During model training, at times, additional metadata such as location is taken into account to train the model which further improves the model accuracy.
5. Select the Routes. These are the routes for which the triage field is configured.
6. Add Field Names.
   1. If the Triage model is trained on a custom field, input the custom field name here. If the model respects

Heuristics Rules

1. Start Values List
   Filter training data on initial custom_data values to ensure training data only includes tickets in scope (ie. starting assignment_group).
   1. Example: assignment_group IN ('value 1', 'value 2', 'value 3')
2. 'Where' conditions
   Add extra filters to ensure training data only includes tickets in scope (ie. specific contact types).
   1. Example: contact_type IN ('Email', 'Self-service', 'Chat', 'ChatBot')
3. Click on Submit

With these steps, you can define the model retraining configurations.