ilyali034/rubert-emotion-ru

Russian Emotion Classifier

A multi-label emotion classifier for Russian texts.
Trained on the CEDR dataset with weighted BCE loss to handle class imbalance.

Base model: cointegrated/rubert-tiny2
F1-micro: 0.7247 | F1-macro: 0.6823

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="ilyali034/rubert-emotion-ru",
    return_all_scores=True,
)

results = classifier("Я очень рад, но немного боюсь")[0]
for r in sorted(results, key=lambda x: -x["score"]):
    if r["score"] > 0.5:
        print(r["label"], round(r["score"], 3))

Example output:

joy 0.892
fear 0.678

Metrics

Metric	Value
F1 micro	0.7247
F1 macro	0.6823
F1 weighted	0.7389
Precision micro	0.6193
Recall micro	0.8733

Per-class F1

Class	F1
joy	0.8391
sadness	0.8057
surprise	0.6711
fear	0.6488
anger	0.4468

Labels

joy · sadness · surprise · fear · anger

Citation

If you use this model in your research, please cite:

@dataset{cedr_v1,
  author = {SAGTeam},
  title = {CEDR: Russian Emotion Dataset},
  year = {2023},
  url = {https://huggingface.co/datasets/sagteam/cedr_v1}
}