Zero-Shot Text Classification in H2O Cascade
- Predictive AI
- H2O Cascade
This lesson walks through creating a new text classification annotation task on a dataset of Amazon reviews. It covers naming the task, selecting the dataset and text column, and defining classes, in this case positive and negative, for the model to predict against. With zero-shot learning enabled by default, the tool assigns a confidence percentage to each comment without any prior training on these specific labels, and the lesson shows how to review and approve at least ten of those predictions before exporting the approved samples as a zip file or to H2O Drive.
Key takeaways
- How to configure a new annotation task by naming it, selecting a dataset, choosing classification, and picking the text column
- What zero-shot learning does here: it predicts a label for a category the model has never specifically been trained on
- How each comment gets a confidence percentage for positive or negative after refreshing the zero-shot predictions
- The keyboard shortcut for quickly approving a prediction you agree with while reviewing samples one by one
- How approved samples and zero-shot predictions can be exported: as a downloadable zip file or directly to H2O Drive
Start owning your AI