metadata
language:
- ar
datasets:
- HARD
tags:
- HARD
widget:
- text: جيد. المكان جميل وهاديء. كل شي جيد ونظيف
- text: استغرب تقييم الفندق كخمس نجوم”. لا شي. يستحق
BERT-ASTD Balanced
Arabic version bert model fine tuned on Hotel Arabic Reviews dataset from booking.com (HARD) dataset balanced version to identify sentiments opinion in Arabic language.
Data
The model were fine-tuned on ~93000 book reviews in arabic using bert large arabic
Dataset:
- Train 70%
- Validation: 10%
- Test: 20%
Results
| class | precision | recall | f1-score | Support |
|---|---|---|---|---|
| 0 | 0.9733 | 0.9547 | 0.9639 | 10570 |
| 1 | 0.9555 | 0.9738 | 0.9646 | 10570 |
| Accuracy | 0.9642 | 21140 |
How to use
You can use these models by installing torch or tensorflow and Huggingface library transformers. And you can use it directly by initializing it like this:
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model_name="mofawzy/Bert-hard-balanced"
model = AutoModelForSequenceClassification.from_pretrained(model_name,num_labels=2)
tokenizer = AutoTokenizer.from_pretrained(model_name)