Table of Contents

Building models in text predictions

After you add new training data for topics or entities that you want to detect in incoming messages, your text prediction contains pending training data. You need to build the topic and entity extraction models to train them with this new training data.

The following example shows the pending training data for entity extraction models: three items of data for the airlines_entities model, and one item for the pySystemEntities model.

Pending training data
The system builds the models in an asynchronous process by using a job scheduler. The job scheduler for building models uses the System Runtime Context (SRC) for rule resolution. To enable the model building process, you must add your application to the SRC.
Add your application to the SRC. For more information, see Automating the runtime context management of background processes.
  1. Open the text prediction:

    1. In the navigation pane of App Studio, click Channels.

    2. In the Current channel interfaces section, click the icon that represents a channel for which you want to configure the text prediction.

    3. On the channel configuration page, click the Behavior tab, and then click Open text prediction.

  2. In the Prediction workspace, click Build.

  3. In the Build models window, select the models that you want to build.

    Do not use the Build all models option. This option builds all default NLP models, such as pySystemEntities and pySmallTalk, and moves them into your application ruleset, which makes it impossible to upgrade these models. This option will be removed from Pega Platform, starting with the 8.6.1 patch release.
    Selecting a model for a new build
  4. Click Build.

    The training of the model starts. The status of the training is displayed on the message bar below the prediction header. You can view detailed progress information by clicking training jobs. At any time, you can stop the model building process by clicking Cancel all builds.
    Model training in progress a text prediction
  5. After the build is complete, click View report to display a summary of the training.

    In the model training report, you can compare the f-scores of the models before and after the training. In the following example, the f-score of the model improved by 0.03 points after the training.
    Model training report

The following video shows a sample model build in a text prediction:

    Related Content

    Have a question? Get answers now.

    Visit the Collaboration Center to ask questions, engage in discussions, share ideas, and help others.