Using machine learning to detect violations in news articles
Haytham Mones from Akhbarmeter, Egypt, has been focusing the majority of his work on tackling manipulation, fake news and disinformation in Egyptian news articles.
He opened the session with a story about recent release of a new AI algorithm called the GPT. One can feed it with one sentence saying for instance the tragic accident that happened in
China with devastating consequences and then the machine learning algorithm generates an entire fake story about this that is coherent and convincing.
Haytham’s work is based on monitoring articles that make it to the top news in Egypt. Every day, he together with his team visit websites that portray top news in Egypt and answer 22 methodological questions regarding human rights violations, manipulation and correctness of information, as well as professional mistakes. Then the articles are awarded a score on a scale from 0-100% based on the answers to these questions. Every month, using this method, they rank newspapers in terms of reliability – from the best to the worst.
They have gathered a large database of articles which shows that over 60% of articles contain violation; 90% of such articles contain human rights violations, while 2% is attributed to fake news. The problem they are facing is a small team: there are only 3 reviewers. Therefore, they have figured out a way to optimize the time they have at hand in order to facilitate efficiency and accuracy by using algorithms for text classification. By quantifying the number or frequency of occurrence of words in an article, they can make mathematical conclusions on the probability that such a document has similar words to a class of other documents. Those classes can be spam, non spam, articles that contain human rights violations or not, etc.
This research could potentially serve to provide a solution for the problems of violations in news articles. The results of the research represent a spotting and publishing of the violations, thus warning the Egyptian public to not read these news articles. The less that unreliable news platforms are used, the more contribution is made towards minimizing the spread of violations in news articles.
Watch the full video: