Zero Shot Text Classification with BERT

Zero Shot Text Classification

This example show how Zero Shot Classification node can be used in order to assign the user provided labels to the texts. No training is required to use Zero Shot classification model. User is only responsible to provide meaningful hypothesis and labels.

In this workflow we compare the predicted values with initial labels, however in practice Zero Shot classification is used in case when labeling the texts is very expensive. So this estimation is only done to demonstrate the performance of the model that has never been trained on the provided or similar data.

External resources

• Benchmarking Zero-shot Text Classification: Datasets, Evaluation and Entailment Approach