mms
MMS (Brand24): 79 sentiment datasets in 27 languages, one config per language.
Labels are the source's -1/0/1 shifted to negative/neutral/positive; original_dataset, domain and the
cleanlab self-confidence are kept. MMS ships a single split, so rows are split 90/5/5 by a hash of the text.
The bundled datasets keep their own licenses; check each original_dataset before use.
Original data: Brand24/mms. Repackaged as parquet for tasksource by scripts/upload_repackaged.py.
- Downloads last month
- -