DarijaMind-Evaluative

Darija Emotion Recognition Model

Model Description

This model is fine-tuned for emotion recognition in Moroccan Darija text.

The architecture is based on DarijaBERT and is designed to classify user messages into emotion categories and emotional intensity levels.

Base Model

• Base model: DarijaBERT-Mix
• Fine-tuning task: Emotion Classification
• Framework: PyTorch
• Library: Transformers

Labels

Emotion Classes

Intensity Levels

Dataset

The model was trained on a Moroccan Darija emotion dataset collected from social media conversations and manually annotated.

Dataset Statistics

Training Details

Hyperparameters

• Learning Rate: 2e-5
• Batch Size: 16
• Epochs: 5
• Optimizer: AdamW
• Max Sequence Length: 128

Regularization

• Dropout
• Weight Decay
• Early Stopping

Evaluation Results

Emotion Classification

Intensity Classification

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "USERNAME/MODEL_NAME"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "كنحس براسي فرحان بزاف اليوم"

inputs = tokenizer(
    text,
    return_tensors="pt",
    truncation=True,
    max_length=128
)

with torch.no_grad():
    outputs = model(**inputs)

prediction = torch.argmax(outputs.logits, dim=-1)

print(prediction.item())

Limitations

• Performance may decrease on dialects outside Moroccan Darija.
• The model can inherit biases from the training data.
• Emotion detection remains challenging for sarcasm and ambiguous contexts.

Citation

@article{elguareh2026darijaemotion,
  title={Emotion Recognition in Moroccan Darija using Transformer-Based Models},
  author={El Guareh, Ayat Allah and others},
  year={2026}
}

Authors

Ayat Allah El Guareh