ArSciKids
A curriculum-aligned dataset for Arabic primary-school science education, comprising two subsets, science experiment records and multiple-choice questions. It covers grades 1 through 6, with content drawn from official textbooks and educational platforms.
Timeline (last 90 days)
ArSciKids contains two independent but complementary datasets: a science experiment dataset and an evaluation question dataset.
The science experiment dataset contains 999 manually curated records from official primary school science textbooks of Saudi Arabia, Syria and Jordan.
The evaluation question dataset contains 2,167 multiple-choice questions collected from Saudi Arabia's Manhaji education platform.
Both datasets cover grades 1 through 6.
Both datasets consist entirely of authentic Arabic educational content rather than machine-translated or synthetically generated material.
The research team fine-tuned ten open-source models using LoRA, with parameter sizes ranging from under one billion to several billion.
Evaluation uses automatic metrics (ROUGE and BLEU) and answer accuracy on multiple-choice generation, supplemented by a five-criteria human evaluation protocol.
The ArSciKids dataset will be publicly available.
The data sources are concentrated in three countries (Saudi Arabia, Syria and Jordan), curriculum coverage is limited, and whether the dataset can generalize to the curricula of other Arabic-speaking countries still requires further verification.
The total of 2167 multiple-choice questions and 999 experimental records is still insufficient for training a large-scale generative model, and is better suited as an evaluation benchmark or fine-tuning seed.
All Facts (10)
The total of 2167 multiple-choice questions and 999 experimental records is still insufficient for training a large-scale generative model, and is better suited as an evaluation benchmark or fine-tuning seed.
50%UnverifiedThe science experiment dataset contains 999 manually curated records from official primary school science textbooks of Saudi Arabia, Syria and Jordan.
50%UnverifiedThe evaluation question dataset contains 2,167 multiple-choice questions collected from Saudi Arabia's Manhaji education platform.
50%UnverifiedBoth datasets cover grades 1 through 6.
50%UnverifiedBoth datasets consist entirely of authentic Arabic educational content rather than machine-translated or synthetically generated material.
50%UnverifiedThe research team fine-tuned ten open-source models using LoRA, with parameter sizes ranging from under one billion to several billion.
50%UnverifiedEvaluation uses automatic metrics (ROUGE and BLEU) and answer accuracy on multiple-choice generation, supplemented by a five-criteria human evaluation protocol.
50%UnverifiedArSciKids contains two independent but complementary datasets: a science experiment dataset and an evaluation question dataset.
50%UnverifiedThe data sources are concentrated in three countries (Saudi Arabia, Syria and Jordan), curriculum coverage is limited, and whether the dataset can generalize to the curricula of other Arabic-speaking countries still requires further verification.
50%UnverifiedThe ArSciKids dataset will be publicly available.
50%