Classify Cantonese/Chinese messages into emotional categories using manual input or file upload.
| Model: | Fine-tuned mBERT (Cantonese/Chinese focus) |
|---|---|
| Training Data: |
|
| Performance: |
Macro F1: 0.89
Accuracy: 89%
|
| Category | Precision | Recall | F1-Score |
|---|---|---|---|
| Academic Stress (學習壓力) | 0.94 | 0.89 | 0.91 |
| Social Isolation (社交孤立) | 0.83 | 0.86 | 0.84 |
| Helplessness (無力感) | 0.84 | 0.81 | 0.82 |
| Experiencing Harm (受到傷害) | 0.93 | 0.93 | 0.93 |
| Positive Sentiment (積極) | 0.92 | 0.96 | 0.94 |
# Training Parameters: k-fold = 5 epochs = 5 batch_size = 16 learning_rate = 3e-5 max_seq_length = 128 lr_scheduler = "cosine" warmup_steps = 500 weight_decay = 0.2 seed = 42