6 Emotions: DistilBERT
Fine-tuned distilbert-base-uncased for seven-class emotion classification:
anger, fear, joy, love, sadness, surprise, and neutral.
Training data
The base data combines EmotionRecognitionDataset, GoEmotions, and AACIntent. AACIntent is stratified to 20,000 rows, text is deduplicated, and each base label is capped at 6,000 rows before a 60%/20%/20% split. Unique SynthSarcasm examples are independently shuffled and stratified 60%/20%/20%, then appended to the matching base splits. AAC scores below 0.35 are labeled neutral.
Test results
- Accuracy: 0.7918
- Macro precision: 0.7926
- Macro recall: 0.7942
- Macro F1: 0.7932
- Weighted precision: 0.7912
- Weighted recall: 0.7918
- Weighted F1: 0.7913
Usage
from transformers import pipeline
classifier = pipeline("text-classification", model="haydenpham/6emotions-distilbert")
classifier("I am so happy to see you.")
- Downloads last month
- 24