Микросервис для бинарной классификации текстовых сообщений (Spam / Ham), обученный на ruSpamModels/russian-spam-detection. За основу взята комбинация TF-IDF + Logistic Regression
- Precision: 0.98
- Recall: 0.82
- F1-score: 0.9
| Модель | Precision | Recall | F1-Score | FP | FN |
|---|---|---|---|---|---|
| Multinomial Naive Bayes (Baseline) | 0.9612 | 0.8594 | 0.91 | 2 407 | 9 760 |
| Logistic Regression | 0.9671 | 0.8976 | 0.93 | 2 121 | 7 107 |
| Порог | Precision | Recall | FP | FN |
|---|---|---|---|---|
| 0.50 (Default) | 96.71% | 89.76% | 2 121 | 7 107 |
| 0.70 | 98.11% | 86.62% | 1 157 | 9 290 |
| 0.85 | 98.88% | 82.52% | 647 | 12 131 |
| 0.90 | 99.18% | 79.97% | 457 | 13 902 |
docker compose up --build
- Web UI: http://localhost:8501
- Swagger: http://localhost:8000/docs