MindLens Emotion Classifier (RoBERTa, dair-ai/emotion)
Fine-tuned roberta-base for 6-class emotion classification
as part of the MindLens mental-health support system.
Classes: sadness, joy, love, anger, fear, surprise
Results (held-out test set, n=2000)
| Metric | Score |
|---|---|
| Accuracy | 0.9250 |
| Macro F1 | 0.8827 |
| 80% gate | PASS |
Dataset
dair-ai/emotion — ~16k English Twitter messages labelled with 6 basic emotions. Train: 16,000 | Val: 2,000 | Test: 2,000
Usage
from transformers import pipeline
clf = pipeline("text-classification",
model="AmiruMallawarachchi/mindlens-emotion-classifier")
clf("I feel completely overwhelmed and hopeless")
Limitations
Twitter-domain training data. One signal among five in the MindLens orchestrator — not a clinical mood assessment.
- Downloads last month
- 6