Wrapper for few typical use cases
Great for semantically distinct classes. Terrible otherwise.
Classifier has been tested on BANKING77 - Dataset composed of online banking queries annotated with their corresponding intents.
| Example Query | Intent |
|---|---|
| I am still waiting on my card? | card_arrival |
| I think my card is broken | card_not_working |
Running example script banking77.py should return precision, recall and f1-score for each label and following micro and weighted averages
| precision | recall | f1-score | support | |
|---|---|---|---|---|
| macro avg | 0.92 | 0.88 | 0.90 | 3072 |
| weighted avg | 0.93 | 0.89 | 0.91 | 3072 |
Label Errors in BANKING77 - Cecilia Ying, Stephen Thomas (https://aclanthology.org/2022.insights-1.19/) highlights a problem with dataset, but it's still a good showcase for this specific use case.