Tackling the challenge of fake news on social media

A Federation University researcher has developed a tool in the fight against online misinformation, with a machine-learning application designed to detect fake news circulating on social media platforms.

Dr Jannatul Ferdush, from Federation’s Centre of Smart Analytics research centre, says the tool was developed because fake news has become a significant challenge on social platforms, often causing unwanted tension and increasing uncertainty about what is real.

“Fake news is constantly evolving, which makes the process of identifying it all the more challenging,” she said.

The research, which focused on a specially curated detection system that identifies fake news through text, was developed as part of her PhD project. Dr Ferdush also experimented with images for the project.

The innovative method for detecting fake news integrates content and comments from multiple social media platforms, aggregating and analysing user interactions from diverse sources to provide a more comprehensive view of the information landscape.

The process involves continually ‘feeding’ the machine-learning tool fake and real news, training the detection system to identify what is not real.

She says the negative impact of fake news underscores the urgency of developing more effective detection systems.

“In fake news, there is often no editorial body, especially on social media. Normally, fake news does not follow grammar correctly, or it has many grammatical mistakes. But with real news, it has an editorial body and mostly very good grammar,” Dr Ferdush said.

“Also, if users from multiple platforms say that something is a real news item, then there is a high chance it is real. This corresponds with the process of identifying fake news from user comments on multiple social media platforms.”

Dr Jannatul Ferdush says fake news is also spreading at an ever-increasing rate, making it harder for detection systems to keep up. As a result, these detection systems will continuously need to be adjusted.

“We need to keep retraining these models of detection because fake news is always evolving, along with the fact that machine learning is trained on historical and not current data,” she said.

“If there is a data drift [when real-world data changes, making a machine learning model less reliable over time], we need to calculate the data drift. If current data is different to historical data, we need to retrain it. This is true for most machine learning cases.”

Dr Ferdush says she hopes to continue developing the tool, which has taken three years to complete, and includes real-life data for training and testing.

Dr Ferdush, who is currently working part-time at Federation, says she is thankful to her supervising team who helped her on her PhD journey and hopes to inspire the next generation by teaching programming and IT in schools.

“Doing a PhD is not always a smooth ride, and you’ll find that a lot of the time things won’t go your way,” she said.

“But, at the end, it’s all worth it. My advice to anyone undertaking a PhD is just to keep going.”