Text classification
IntentRoute
Intent recognition across 77 banking categories.
Explore report →
Recorded study · report data

Study results
77 categories · English test
- Macro-F1 · English test
- 90.74%
- Polish diagnostic examples
- 12
Method
A frozen multilingual MiniLM encoder and trained MLP classifier, compared with TF-IDF.
- Input
- Request text
- Output
- Category and model score
Limits of the result
Accuracy on 12 authored Polish examples: 66.67%. This is not a representative Polish benchmark.