Researchers have developed an automated system for identifying fake news that obtained accuracy rate above 98% in tests on two datasets; in other words, it missed just 1 or 2 fake news items in every 100. They offer information of their strategy within the International Journal of Information and Computer Security.
The new system uses a transformer encoder, an artificial intelligence (AI) model that detect relationships between phrases and their context. This is integrated with a deer-hunting optimization algorithm (DHOA), a computational method for finding effective model settings, based on modeling the way wherein human hunters track and pursue prey. The researchers removed web addresses, punctuation and common, or “stop,” words before applying their analysis to the text. The DHOA component automatically fine-tunes the model’s settings instead of requiring manual selection and calibration.
Automated detection could accelerate manual fact-checking, that is frequently costly and time-consuming, permitting social media platforms to display screen out misleading claims quickly to avoid their rapid spread.
The next step might be to test the system on real-world social media content to see whether it can distinguish deceptive material from the legitimate reporting it intently resembles. As it stands, the approach is restrained to text, instead of checking for the manipulation of images, deepfake videos or inconsistencies between text and visual evidence.
Future work might discover how the ones different media types might be processed and analyzed as well as searching out for relationships between content and the users who post it as important clues as to whether a given item is fake news or not.
