Training the NLP model
When you create a Voice AI channel, Pega Customer Service generates a text prediction. The text prediction includes the topic, sentiment, and entity models that Voice AI uses to identify topics and entities in a customer conversation.
This demo shows how to help train these models by adding sample data.
Transcript
In this demo, we add data to the Voice AI channel to help train the NLP model to identify a customer home address.
Log in as a customer service administrator.
In the Voice AI channel, click the Training data tab.
To add the training data, add records that represent requests from customers. The records are associated with the Account address change topic.
Review a record to see the classification details.
The NLP analysis section shows the language model, topic, and any entities detected. Above the analysis section, you see the text that represents the customer conversation. The detected entities in this text are highlighted.
Review the remaining records, and then adjust them as needed.
Click Select All to select all the records.
Click Mark reviewed to add the records to the build queue.
Select Build Model to rebuild the NLP model, which adds the new records and updates the model.
If the model build succeeds, you receive a message that shows the new F-Score at the top. The F-Score measures the effectiveness of the model.
In the Voice AI channel, click the Behavior tab, then click Open Text Prediction.
In Prediction Studio, you can manage all aspects of the NLP models, including adding training records.
Click the Training tab. You can add and review records and build the model.
You've reached the end of this video.
This Topic is available in the following Module:
Want to help us improve this content?