Creating Intent Datasets

Models are trained using a large set of customer messages called data sets. To create data sets, the user will go through a group of messages and label (or categorize) them with the Intent category that they belong to. For example, if a user wants to add an intent for "Lunch Break Time", they would take a group of messages from their system, label them with the Intent label of "Lunch Break Time" and train the model on that data set.

The amount of messages required to create a quality model various greatly dependent on the use case, and how much variation could be used when asking questions. For examples, a user might send a message that has the same Intent but is worded differently each time.

Each Intent should initially be trained on data sets that vary in size, dependent on the Intent complexity and variance of general phrasing. After the model is tested, the user can continue to add new data to train the model on to refine the results.

The Labeling Tool is an easy way to quickly label messages that can be used for creating data sets. The Connection with the users messaging system allows for a user to jump right in and begin to quickly label each message by simply clicking on one of the Intent "pills" in the right column and cycling to the next message.