1 Select Task Type
Which task should I pick?
- Classification: assign one or more labels to each recommendation.
- Text extraction: same workflow, but treated as detecting whether target concepts appear in text.
🏷️
Multi-Label Classification
Assign categories to text
2 Define Categories
Add as many categories as needed. The special category excluded is auto-added and learned by the model, but ignored in statistics.
3 Label Training Samples
0 / 0 labeled
• 0 excluded
Define categories once, then for each sample select only the categories that apply. Different samples can have different category combinations.
4 Train & Apply
Simple recommendation:
- Start with TextSetFit (recommended) and label at least 15-20 samples.
- Check benchmark scores before applying to all records.
- If precision is low, increase threshold; if recall is low, decrease threshold.
Classification method
Balanced precision/recall with sparse text vectors.
Checking SetFit backend...
Prediction threshold
Lower: more labels (higher recall). Higher: fewer labels (higher precision).
0.15
Training in progress...
0%
Benchmark (Hold-out Test Data)