SAMO Emotion Classifier (multi‑label)

Multi‑label emotion detection for journal‑style text. Trained on GoEmotions and adapted for real‑world journaling. Outputs a probability for each emotion and a primary label.

Intended use

• Emotion analysis of short texts/journal entries
• UX insights, content moderation, well‑being apps
• Not a medical device; do not use for diagnosis or high‑risk decisions

Labels

The model predicts the GoEmotions set (27 emotions + neutral). Ensure id2label/label2id are present in the config.

How to use

from transformers import pipeline
clf = pipeline(
    "text-classification",
    model="0xmnrv/samo",
    tokenizer="0xmnrv/samo",
    top_k=None,           # return all labels with scores
    truncation=True
)
texts = [
    "I am feeling really happy today!",
    "I'm frustrated but hopeful."
]
print(clf(texts))

Or with logits/sigmoid manually:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch, torch.nn.functional as F

tok = AutoTokenizer.from_pretrained("0xmnrv/samo")
mdl = AutoModelForSequenceClassification.from_pretrained("0xmnrv/samo")
x = tok("I am feeling really happy today!", return_tensors="pt", truncation=True)
with torch.no_grad():
    p = torch.sigmoid(mdl(**x).logits).squeeze(0)
id2label = mdl.config.id2label
scores = {id2label[i]: float(p[i]) for i in range(len(p))}
print(scores)

Training & data

• Base model: roberta-base
• Dataset: GoEmotions (+ optional domain data if applicable)
• Objective: multi‑label classification (sigmoid + BCEWithLogits)
• Preprocessing: 512 max tokens, lowercase normalization

#Evaluation #- Report macro‑F1 and micro‑F1 on validation/test #- Example (replace with your numbers): #- Macro F1: #- Micro F1: #- Accuracy:

Limitations & bias

• May struggle with sarcasm/irony and long context
• Emotion distributions reflect GoEmotions and any domain data used
• Not suitable for clinical use

Safety & privacy

• Do not send PII unless you comply with your jurisdiction’s rules
• For sensitive content, prefer private repos and authenticated endpoints

Citation

If you use this model, please cite:

• GoEmotions: https://arxiv.org/abs/2005.00547
• Transformers: https://github.com/huggingface/transformers