qarib/bert-base-qarib_far_9920k

QARiB: QCRI Arabic and Dialectal BERT

About QARiB Farasa

QCRI Arabic and Dialectal BERT (QARiB) model, was trained on a collection of ~ 420 Million tweets and ~ 180 Million sentences of text. For the tweets, the data was collected using twitter API and using language filter. lang:ar. For the text data, it was a combination from Arabic GigaWord, Abulkhair Arabic Corpus and OPUS. QARiB: Is the Arabic name for "Boat".

Model and Parameters:

• Data size: 14B tokens
• Vocabulary: 64k
• Iterations: 10M
• Number of Layers: 12

Training QARiB

See details in Training QARiB

Using QARiB

You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to be fine-tuned on a downstream task. See the model hub to look for fine-tuned versions on a task that interests you. For more details, see Using QARiB

This model expects the data to be segmented. You may use Farasa Segmenter API.

How to use

You can use this model directly with a pipeline for masked language modeling:

>>>from transformers import pipeline
>>>fill_mask = pipeline("fill-mask", model="./models/bert-base-qarib_far")
>>> fill_mask("و+قام ال+مدير [MASK]")
[

]
>>> fill_mask("و+قام+ت ال+مدير+ة [MASK]")
[
]
>>> fill_mask("قللي وشفيييك يرحم [MASK]")
[
]

Evaluations:

Experiment	mBERT	AraBERT0.1	AraBERT1.0	ArabicBERT	QARiB
Dialect Identification	6.06%	59.92%	59.85%	61.70%	65.21%
Emotion Detection	27.90%	43.89%	42.37%	41.65%	44.35%
Named-Entity Recognition (NER)	49.38%	64.97%	66.63%	64.04%	61.62%
Offensive Language Detection	83.14%	88.07%	88.97%	88.19%	91.94%
Sentiment Analysis	86.61%	90.80%	93.58%	83.27%	93.31%

Model Weights and Vocab Download

From Huggingface site: https://huggingface.co/qarib/bert-base-qarib_far

Contacts

Ahmed Abdelali, Sabit Hassan, Hamdy Mubarak, Kareem Darwish and Younes Samih

Reference

@article{abdelali2021pretraining,
    title={Pre-Training BERT on Arabic Tweets: Practical Considerations},
    author={Ahmed Abdelali and Sabit Hassan and Hamdy Mubarak and Kareem Darwish and Younes Samih},
    year={2021},
    eprint={2102.10684},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}