AI detects fake Arabic news with 94% accuracy

STAFF REPORTER
MUSCAT, DEC 20
A team at Sultan Qaboos University, led by Dr Ahmed Shahata from the College of Arts and Social Sciences, has developed a groundbreaking approach to tackling fake news in Arabic media, including social networks.
The team introduced ArabFake, a deep-learning algorithm designed to detect misleading news, classify content and evaluate the potential risks posed by its spread.
Built on the advanced MARBERTv2 model for multi-dialect Arabic tweets, ArabFake addresses the complexities of the Arabic language while performing three critical tasks: identifying fake news, categorising content and assessing risk.
The algorithm was trained on a verified dataset of 2,495 news items, labelled by experts for authenticity and risk, and tested on two large datasets — ANS Corpus and AraNews — with nearly 200,000 Arabic news articles, both genuine and fabricated.
The results were impressive. ArabFake achieved 94.12 per cent accuracy in detecting fake news, 84.92 per cent in content classification and 88.91 per cent in risk assessment, demonstrating its reliability across multiple tasks.
The study also revealed patterns in Arabic fake news: fabricated stories accounted for 60.4 per cent of the dataset, while misleading economic information made up 22.4 per cent. Nearly two-thirds of fake news was considered highly risky for society, highlighting the urgent need for effective detection systems.
ArabFake’s innovative use of equity recording techniques allowed it to identify linguistic patterns associated with fake news, providing insights into misinformation trends.
By simultaneously evaluating content authenticity, estimating risk levels and prioritising interventions, the algorithm offers practical opportunities for news organisations, fact-checking initiatives, content-moderation systems and media literacy programmes.