ArabFactCheck: An Arabic Fake News Dataset with Fine-Tuned Language Models for Misinformation Detection

The dataset contains 7,691 Arabic news instances written primarily in Modern Standard Arabic and is designed to support research in natural language processing, misinformation detection, fact-checking, and explainable AI. Each fake news instance is accompanied by verification text that explains the false claim and provides supporting evidence.

The news records span approximately February 2016 to July 2025.</p><p dir="ltr">Each record in the dataset is assigned a label, either real or fake. The distribution of real and fake samples is as follows:</p><ul><li>4,025 real news articles</li><li>3,666 fake news articles</li></ul><p dir="ltr">The arabfactcheck_data.jsonl file contains the complete dataset of 7,691 records. The split folder contains the corresponding training, validation, and testing splits, which can be used for model development and evaluation.</p>