New Tool Developed at Concordia Targets Social Media’s Fake News Problem

In a new approach to combating misinformation online, researchers have developed “SmoothDetector,” an advanced model that offers a more nuanced method of identifying fake news on social media platforms by examining multiple content types simultaneously.

According to Adegboyega Ojo, a key researcher behind the project, SmoothDetector represents an evolution in fake news detection technology. Unlike earlier systems that could only assess one medium at a time—either text, images, audio, or video—this new model analyzes all elements of a post together, providing a more comprehensive evaluation.

“Previous models had significant limitations,” Ojo explained. “A post containing false text but an accurate image could be misclassified, leading to false positives or negatives in the detection process.”

This advancement is particularly valuable during breaking news events, when information flows rapidly and often contains contradictory elements. The traditional binary classification of content as either “real” or “fake” fails to capture the complex nature of online information.

“We wanted to capture these uncertainties to make sure we were not making a simple judgement on whether something was fake or real,” Ojo noted. “This is why we are working with a probabilistic model. It can monitor or control the judgement of the deep learning model. We don’t just rely on the direct pattern in the information.”

The system’s name derives from its core technical approach—smoothing the probability distribution of outcomes. Rather than making definitive declarations about content authenticity, SmoothDetector evaluates inherent uncertainties in the data and quantifies likelihood, resulting in more sophisticated assessments.

“This makes it more versatile to capture both positive and negative information or correlation,” added Ojo, highlighting the model’s ability to handle ambiguous cases that would challenge conventional systems.

The technology has been specifically tested on Twitter (now X) and Weibo, two major social media platforms with different user bases and content patterns. Weibo, often described as China’s equivalent to Twitter, provides researchers with valuable comparative data from a different cultural and linguistic context.

While the current version of SmoothDetector focuses primarily on text and image analysis, Ojo indicated that further development is underway to incorporate audio and video processing capabilities. This would transform it into a truly comprehensive multimodal system capable of analyzing all aspects of digital content across various social media environments.

The research represents a collaborative international effort, with contributions from Nizar Bouguila, a professor at the Concordia Institute for Information Systems Engineering, along with assistant professor Fatma Najar from John Jay College of Criminal Justice, and assistant professors Nuha Zamzami and Hanen Himdi from the University of Jeddah in Saudi Arabia.

This development comes at a critical time in the fight against digital misinformation. Social media platforms have faced mounting pressure from regulators, users, and advertisers to address the spread of fake news, which has been linked to real-world consequences ranging from public health crises to political instability.

Traditional content moderation approaches have struggled to keep pace with the volume and sophistication of misleading content online. Automated systems often lack the nuance to distinguish between deliberate misinformation and legitimate content with unusual characteristics, while human moderation faces scalability challenges.

SmoothDetector’s probabilistic approach represents a potentially significant advancement in addressing these challenges by moving beyond simple binary classifications to recognize the inherent complexity of online information.

The research team’s findings have been published in the paper “SmoothDetector: A Smoothed Dirichlet Multimodal Approach for Combating Fake News on Social Media,” available through IEEE Explore, a leading repository for technical literature in engineering and technology.