What used to required a specialized model trained on task-specific labeled data can now be achieved easily with Airtrain AI.

Whether you are looking to classify your dataset along a taxonomy specific to your business logic, or do a binary sentiment analysis, Airtrain's classification feature lets you do that really easily.

Ingestion and configuration

As an example, let's say you have a dataset of smart assistant commands (e.g. Alexa) that you want to classify across themes such as reminder, weather, news, messaging, timer, music, news, recipes, etc.

On the dataset upload form, check the box for "AI classification & labeling" and list the classes of interest. For each class, specify a name, and a meaningful description. You can add as many classes as you want.

For optimal results, make sure the description are as... descriptive as possible to allow our classification engine to easily identify a row's class and differentiate it from other classes.

Once you have described all classes, click "Create dataset" to trigger ingestion and classification.

Visualization and filtering

Once the dataset has completed ingestion and classification, you can visualize the class breakdown in the corresponding pie chat.

You can then click on any of the pie chart slices to filter your dataset accordingly.

Export

When exporting your dataset to JSONL, the generated classification will be included in the exported data.