A Probabilistic Reasoning Framework to Detect Fake News on Social Media

The rapid growth of social media has made the Internet a critical platform for spreading misinformation, which shapes public opinion and harms society. Despite significant research in fake news detection, most probabilistic efforts rely heavily on Naive Bayes, with limited exploration of other probabilistic models. This paper introduces a Bayesian network (BN) modelling-based framework for fake news detection, offering a probabilistic estimate of the likelihood of news being false. Unlike binary classification, this approach reflects human decision-making by evaluating three key questions: “who”, “what”, and “when”. Each module corresponds to a specific feature set, enabling nuanced reasoning about the news’s credibility. The framework is flexible, allowing adjustments through expert input, knowledge bases, or real-world data. We validate the approach using a semi-synthetic dataset containing features from news content, user behaviour, and social context. The results highlight the framework’s capacity to leverage expert knowledge, providing a more reliable and adaptive solution compared to traditional classifiers. The BN-based method demonstrates enhanced robustness, positioning it as a promising tool for tackling misinformation in an evolving digital landscape.