SkyWater21/lvbert-lv-go-emotions-ekman

Fine-tuned LVBERT for multi-label emotion classification task.

Model was trained on lv_go_emotions dataset. This dataset is Latvian translation of GoEmotions dataset. Google Translate was used to generate the machine translation.

Original 26 emotions were mapped to 6 base emotions as per Dr. Ekman theory.

Labels predicted by classifier:

0: anger
1: disgust
2: fear
3: joy
4: sadness
5: surprise
6: neutral

Label mapping from 27 emotions from GoEmotion to 6 base emotions as per Dr. Ekman theory:

GoEmotion	Ekman
admiration	joy
amusement	joy
anger	anger
annoyance	anger
approval	joy
caring	joy
confusion	surprise
curiosity	surprise
desire	joy
disappointment	sadness
disapproval	anger
disgust	disgust
embarrassment	sadness
excitement	joy
fear	fear
gratitude	joy
grief	sadness
joy	joy
love	joy
nervousness	fear
optimism	joy
pride	joy
realization	surprise
relief	joy
remorse	sadness
sadness	sadness
surprise	surprise
neutral	neutral

Seed used for random number generator is 42:

def set_seed(seed=42):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    if torch.cuda.is_available():
        torch.cuda.manual_seed_all(seed)

Training parameters:

max_length: null
batch_size: 32
shuffle: True
num_workers: 2
pin_memory: False
drop_last: False

optimizer: adam
lr: 0.000005
weight_decay: 0

problem_type: multi_label_classification

num_epochs: 3

Evaluation results on test split of lv_go_emotions

	Precision	Recall	F1-Score	Support
anger	0.57	0.41	0.47	726
disgust	0.70	0.26	0.38	123
fear	0.66	0.55	0.60	98
joy	0.82	0.79	0.80	2104
sadness	0.72	0.46	0.56	379
surprise	0.63	0.49	0.55	677
neutral	0.66	0.61	0.63	1787
micro avg	0.72	0.61	0.66	5894
macro avg	0.68	0.51	0.57	5894
weighted avg	0.71	0.61	0.65	5894
samples avg	0.65	0.63	0.63	5894