Source-Aware Kurdish Fake News Detection: Bridging Qur'anic Wisdom and Deep Learning

In the digital age, information travels faster than ever, but so does misinformation. When false stories spread, they can damage public trust, confuse communities, and even endanger lives. For decades, scientists have tried to build computer programs that can spot these lies automatically. These programs usually work by reading the text of an article and looking for patterns that suggest deception, such as unusual word choices or emotional language. However, this approach has a major blind spot: it often ignores the most important clue of all, the source. Just as a person might trust a warning from a doctor more than a rumor from a stranger, a computer needs to know who is speaking to judge what is being said. This is especially difficult in languages like Kurdish, where there are fewer digital tools and less data to train these programs, making the task of separating truth from fiction even harder.

A team of researchers from Charmo University and the University of Kurdistan Hewler in Iraq set out to solve this problem by building a system specifically for Kurdish news. They started with a massive collection of nearly 101,000 news articles, half of which were verified as real and half as fake. To teach their computer models, they tried two different approaches. In the first approach, the models were fed only the text of the news articles, forcing them to guess whether a story was true based on the words alone. In the second approach, the models were given the text along with the name of the news organization that published it. This second method was inspired by an ancient principle found in the Qur'an, which instructs believers to verify the source of any information before accepting it as truth. The researchers wanted to see if this old wisdom held up in the modern world of artificial intelligence.

The results were striking. When the computer models were allowed to see the name of the news source, they became nearly perfect at their job. The best models, which used deep learning techniques to understand the structure of the language, achieved a near-perfect score, correctly identifying almost every single fake and real news story in their test set, with the top CNN model making only a single error. They did this with incredible speed, taking only ten seconds to train on the data and a fraction of a second to analyze a new article. However, when the researchers removed the source names and asked the models to rely on the text alone, the results changed dramatically. The performance of the best models dropped by nearly eighteen percent. Even the most advanced artificial intelligence struggled to distinguish truth from lies without knowing who was speaking.

This drop in performance was not a small glitch; it was a clear signal that the text of a news story is often not enough to tell if it is fake. The researchers found that traditional computer programs, which use simpler mathematical rules, fell apart even more when they lost the source information, while the more advanced deep learning models held up slightly better but still failed to reach their full potential. The study showed that the difference between a perfect system and a flawed one was simply the inclusion of the publisher's name. This finding confirms that in the Kurdish media landscape, and likely in many others, the credibility of the outlet is just as important as the content of the story.

The researchers concluded that the most effective tool for the future is a specific type of deep learning model called a Convolutional Neural Network, which proved to be both perfectly accurate and incredibly fast. This system is now ready to be used in real-world situations to help fact-checkers and news organizations in the Kurdish-speaking world. More broadly, the study offers a powerful lesson for anyone trying to fight misinformation: you cannot judge a story by its words alone. The source matters. By combining modern technology with the timeless advice to check where information comes from, the researchers have created a system that is not only smarter but also more culturally aware. Their work proves that when we build tools to understand the world, we must include the context of who is speaking, or we risk missing the truth entirely.