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

Area of Science:

• Computational Social Science
• Artificial Intelligence
• Machine Learning

Background:

• Social media's rapid growth presents challenges in information verification.
• Fake news proliferation erodes journalistic credibility and fuels societal discord.
• Automated fake news detection is crucial due to the overwhelming volume of online content.

Purpose of the Study:

• To develop an unsupervised fake news detection method, eliminating the need for manual data labeling.
• To evaluate the effectiveness of clustering algorithms (GMM, K-means, K-medoids) for fake news detection.
• To propose and validate a novel hybrid GMM-GCO model for enhanced fake news identification.

Main Methods:

• Unsupervised learning utilizing clustering algorithms: Gaussian Mixture Model (GMM), K-means, and K-medoids.
• Development of a hybrid GMM approach integrated with the Group Counseling Optimizer (GCO) metaheuristic algorithm.
• Comparative analysis of clustering performance using metrics like silhouette score, ARI, and purity on real-world datasets.

Main Results:

• The proposed hybrid GMM-GCO method demonstrated superior performance compared to existing techniques.
• Achieved high clustering quality scores: silhouette score of 0.77, ARI of 0.83, and purity score of 0.88.
• Effectively eliminated the requirement for manually labeled datasets in fake news detection.

Conclusions:

• The hybrid GMM-GCO approach offers a promising unsupervised solution for accurate fake news detection.
• This method addresses the limitations of supervised learning approaches by removing the need for extensive data annotation.
• The findings indicate a significant advancement in automated fake news detection systems, enhancing information integrity.