nuha

Model Summary

nuha is a lightweight, ONNX-optimised Arabic text classifier that categorises Jordanian social media comments into three classes based on the NUHA methodology for online gender-based violence (OGBV). It fine-tunes nuha-mlm — a domain-adapted Arabic BERT — with a reduced 4-layer architecture for efficient CPU inference, and is exported to ONNX. It shares the same classification task and labels as nuha-multiclass but is optimised for production deployment. This is the model powering the NUHA analysis platform.

Label	Meaning
Not Online Violence	Comments that are not hate speech
Offensive Language	Hate speech characterised by irony or sarcasm
Gender Based Violence	Direct hate speech targeting gender — the primary focus of NUHA

This model was developed as part of a pilot proof-of-concept for the NUHA project by the Jordan Open Source Association (JOSA).

For the full-depth (12-layer) version of this classifier, see nuha-multiclass.

Uses

Direct Use

from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer, pipeline

model = ORTModelForSequenceClassification.from_pretrained("thejosango/nuha")
tokenizer = AutoTokenizer.from_pretrained("thejosango/nuha")
classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)

result = classifier("اخرسي يا غبية")
print(result)
# [{'label': 'Gender Based Violence', 'score': ...}]

For batch inference:

comments = ["يعطيكم العافية", "أنتِ ساحرة", "اخرسي يا غبية"]
results = classifier(comments)
for comment, result in zip(comments, results):
    print(f"{result['label']} ({result['score']:.2f}): {comment}")

Using the PyTorch Version

If you need the full PyTorch model (for fine-tuning or non-ONNX inference), use nuha-multiclass directly.

Out-of-Scope Use

• Other Arabic dialects: The model was trained primarily on Jordanian Arabic. Performance on Egyptian, Gulf, or Modern Standard Arabic is not validated.
• Other hate speech targets: NUHA is calibrated for online gender-based violence. It is not designed to detect hate speech targeting race, religion, or other demographics.
• High-stakes automated decisions: Given the moderate performance (F1 ≈ 0.54) and pilot nature of this work, the model should not be used as the sole decision-maker in content moderation systems without human review.

Preprocessing

At inference time, apply the following normalisation to input text before passing it to the model:

1. URLs replaced with [رابط] token
2. @mentions replaced with [مستخدم] token
3. Email addresses replaced with [بريد] token
4. Numbers removed
5. Punctuation removed
6. Arabic diacritics (harakat) removed
7. Whitespace normalised

Evaluation Results

Evaluated on the validation split of thejosango/nuha-dataset (methodology configuration):

Metric	Value
F1 (macro)	0.5363
Precision	0.6660
Recall	0.5188

See nuha-multiclass for full training details and evaluation discussion.

---

This model was developed as part of an initial pilot study. Performance metrics reflect the complexity of the task and the proof-of-concept nature of this system.