Create a model to predict similar knowledge articles

Use the Task Intelligence for CSM to create and train a machine learning model that recommends relevant knowledge articles based on the context of open incidents. The model analyzes existing data to display relevant articles when working on a current case, helping agents find useful information faster. The plugin includes a ready-to-train model for predicting similar articles and also enables you to create custom models tailored to your data.

Before you begin

• Ensure that the Task Intelligence for Customer Service plugin is installed.
• Ensure that your instance contains sufficient case and knowledge data (minimum 10,000 recommended) for meaningful training.
• Role required: ml_admin, ti_admin

Set up the prediction model

Procedure

1. Navigate to All > Task Intelligence for Customer Service > Setup.
   The Task Intelligence Admin Console displays.
2. On the Similar Knowledge Articles card, select Ready to train.
   The model opens in a guided setup flow. The Define the purpose screen appears.
3. Define the purpose:
   1. The Prediction table and Training table are pre-selected as Cases and Knowledge, respectively.
      These values are fixed and can’t be edited.
   2. Select Save & Continue.
      The Train your model screen appears.
4. Train your model:
   1. The Model name and Prediction table fields are pre-filled and can’t be edited.
   2. Optional: Apply Conditions to filter the training data.
   3. In the Prediction fields field, select Short description that the model uses to identify similarities.
   4. Note:

      The options selected here are defaults. You can modify the options in the Prediction fields and Training fields based on your requirements.
      In the Training fields field, select Article body and Short description of the Knowledge training table.
   5. Choose the language for training.
   6. Set the Update frequency.
   7. Review the Number of records.
      If needed, select the Load records icon to reload.
   8. Optional: Enable Auto-retrain to enable retraining the model on a set schedule.
   9. Select Launch training.
   10. Once training starts, select View current results to preview sample outputs.
       The Assess and define screen appears.
5. Assess and define:
   1. Review the Estimated number of records used for training.
   2. In Similar Knowledge preference, select one of the following options:
      • Recommendations – Recommends similar records based on training and prediction fields (selected by default).
      • Monitor only – Runs the model in the background without displaying recommendations. You can analyze the logged data before enabling recommendations.
      • Turn off predictions – Disables all predictions for this model.
   3. Select Save & Continue.
      The Deploy your model screen appears.
6. Deploy your model:
   1. Review your setup and training summary and then select Deploy to activate the model.
      A confirmation message appears when deployment is complete.
   Result

   Once deployed, similar knowledge articles appear in the Recommended Actions section under the Suggested Actions tab when an agent opens a case, helping them resolve cases faster.