باحثون سعوديون يطورون نظاماً لكشف الأخبار المضللة بدقة 98.23 %

Researchers from the universities of Tabuk and King Saud, in collaboration with a researcher from Arizona State University in the United States, have developed a multimedia artificial intelligence system to detect fake and misleading news during wars and crises, achieving an accuracy of 98.23% in one database related to the Ukraine crisis, and over 94% in four international databases.

The study was published in the journal "Scientific Reports" by "Springer Nature" on October 16, 2026, and highlighted the shortcomings of traditional automated verification systems in dealing with misinformation during crises, as it does not only involve the dissemination of explicit false information but may also rely on the combination of text, images, and emotional charge, making each element appear acceptable on its own, while the misinformation arises from the way they are assembled together.

4 Databases for Testing

The researchers relied on four extensive standard databases used to test misinformation detection techniques: "Fakedet," "Ukraine-GitHub," "Weibo," and "MediaEval," allowing for the evaluation of the system's performance across diverse digital environments in terms of language, platform, and content nature, rather than being limited to a single database.

The system recorded an accuracy of 94.66% in the "Fakedet" database, 98.23% in "Ukraine-GitHub," 98.12% in "Weibo," and 97.98% in "MediaEval." The highest result came from the database related to Ukraine, exceeding the "Fakedet" result by 3.57 percentage points.

3 Analytical Paths for News Analysis

The system relies on three integrated analytical paths, instead of assigning the verification task to a single algorithm. The first path focuses on text analysis using a multilingual language model to extract meaning, context, and relationships between sentences, while the second path examines images using computer vision techniques to extract visual features and understand their content.

The third path analyzes emotions and emotional tone through models trained to detect emotional signals that may be exploited in creating misleading narratives, especially during conflicts, where fear, anger, and polarization are increasingly employed to influence recipients.

This approach becomes particularly important during times of war and crises, characterized by the rapid circulation of content, the density of visual materials, and the difficulty of immediate verification, in addition to the potential use of real materials within misleading contexts. An image may be original, but its description, timing, or association with a specific event may be incorrect, while emotional phrasing may lead the recipient to believe the narrative before verifying its details.

Tests showed that integrating the three paths consistently achieved higher performance compared to a set of modern methods used by the researchers for comparison. Tests excluding some components of the system also confirmed the importance of combining multiple types of information, rather than relying on text or images separately.

Additional Tests for Arabic Content

The study concluded that addressing digital misinformation during wars and crises requires systems that analyze news as an interconnected system of text, images, and emotions, not just isolated words, and that combining specialized models and integrating their results may enhance detection stability and reduce reliance on the weaknesses of a single model.

The results highlight the need to test the system on Arabic content and local dialects, measure its performance on news circulating at the moment events occur, and evaluate its ability to handle images and videos produced or modified using artificial intelligence techniques.