Unsupervised fake news detection on social media using hybrid Gaussian Mixture Model.

The rise of social media has revolutionized information dissemination, creating new opportunities but also significant challenges. One such challenge is the proliferation of fake news, which undermines the credibility of journalism and contributes to societal unrest. Manually identifying fake news is impractical due to the vast volume of content, prompting the development of automated systems for fake news detection. This challenge has motivated numerous research efforts aimed at developing automated systems for fake news detection. However, most of these approaches rely on supervised learning, which requires significant time and effort to construct labeled datasets. While there have been a few attempts to develop unsupervised methods for fake news detection, their reported accuracy results thereof remain unsatisfactory. This research proposes an unsupervised approach using clustering algorithms, including Gaussian Mixture Model (GMM), K-means, and K-medoids, to eliminate the need for manual labeling in detecting fake news. In particular, it also proposes a novel hybrid method that leverages the Gaussian Mixture Model (GMM) in conjunction with the Group Counseling Optimizer (GCO), a metaheuristic optimization algorithm, to identify the optimal number of clusters for the detection of fake news. The comparative analysis of the evaluation results on real-world data demonstrated that the proposed hybrid GMM outperforms the state-of-the-art techniques, with a silhouette score of 0.77, ARI of 0.83, and a purity score of 0.88, indicating a significantly improved quality of clustering results.