SahmBenchmark/arabic-business-mcq_training_standardized

Arabic Business MCQ Training Dataset (Standardized)

Standardized training dataset for Arabic Business MCQ, formatted to match the evaluation splits.

Dataset Description

This dataset contains the training split of the Arabic Business MCQ dataset, standardized to match the format used in the evaluation splits.

Dataset Summary

• Language: Arabic
• Domain: Business, Economics, Entrepreneurship, Accounting
• Format: Multiple choice questions (standardized)
• Task: Business knowledge training

Dataset Structure

Data Splits

• Train: 381 examples (full training dataset)

Data Fields

• id: Unique identifier (e.g., "business_0003")
• query: Full MCQ prompt with instructions in Arabic
• answer: Correct answer letter (a, b, or c)
• text: Question text without instructions
• choices: List of answer options as letters ['a', 'b'] or ['a', 'b', 'c']
• gold: Zero-based index of correct answer

Question Types

The dataset includes:

1. True/False questions: 2 choices (a: صح, b: خطأ)
2. Multiple choice questions: 3 choices

Example

{
  "id": "business_0003",
  "query": "اقرأ بعناية السؤال التالي والإجابات المحتملة...",
  "answer": "b",
  "text": "التكاليف غير المباشرة هي التكاليف التي...",
  "choices": ["a", "b"],
  "gold": 1
}

Usage

from datasets import load_dataset

# Load the standardized training dataset
dataset = load_dataset("SahmBenchmark/arabic-business-mcq_training_standardized")

# Access training data
train_data = dataset['train']

# Training example
for example in train_data:
    # Use example['query'] as input
    # Use example['answer'] as target
    model_input = example['query']
    expected_output = example['answer']
    
    # Or use the gold index
    correct_idx = example['gold']
    correct_choice = example['choices'][correct_idx]

Related Datasets

• Evaluation splits: SahmBenchmark/arabic-business-mcq_eval
• Original dataset: SahmBenchmark/arabic-business-mcq

Citation

@dataset{arabic_business_mcq_training_2025,
  title={Arabic Business MCQ Training Dataset (Standardized)},
  author={SahmBenchmark},
  year={2025},
  publisher={Hugging Face}
}