Muscat – A research team at Sultan Qaboos University (SQU) has developed a deep-learning framework that could significantly strengthen efforts to detect fake news in Arabic media, including on social networking platforms.
Led by Dr Ahmed Shahata from the College of Arts and Social Sciences, ArabFake is an AI-based model designed to identify misleading news, classify its content and assess the potential risks arising from its circulation.
ArabFake is built on MARBERTv2, a model designed to process multiple Arabic dialects commonly used in online content. The framework addresses the linguistic complexity of Arabic while performing three tasks simultaneously: detecting fake news, categorising the type of content and estimating its level of risk.
To develop the model, the researchers trained ArabFake on a verified dataset of 2,495 news items, each labelled by specialists according to authenticity and risk level. The system was then tested on two large Arabic-language datasets – ANS Corpus and AraNews – covering nearly 200,000 news articles, both genuine and fabricated.
The results showed strong performance across all areas. ArabFake achieved an accuracy rate of 94.12% in fake news detection, 84.92% in content classification and 88.91% in risk assessment. The researchers said the findings demonstrate the model’s reliability and its ability to handle multiple analytical tasks at once.
It also highlighted key trends in Arabic misinformation. Fabricated stories accounted for 60.4% of the analysed content, while misleading economic news made up 22.4%. Nearly two-thirds of fake news items were classified as high risk, underlining the potential impact of these on public opinion and social stability.
According to the researchers, ArabFake’s use of advanced linguistic pattern analysis enables it to detect subtle indicators of false content. The system can assess published material in real time, determine its credibility, assign a risk level and help prioritise rapid responses to high-risk misinformation.
The team said the model has practical applications for news organisations, fact-checking platforms, content moderation systems and media literacy programmes, offering a data-driven tool to curb the spread of misleading information and strengthen trust in Arabic-language media.
