Classification of unlabeled online media

Area of Science:

• Social Network Analysis
• Information Science
• Computational Social Science

Background:

• Misinformation detection typically requires labeled datasets, limiting scalability.
• Existing methods often rely on content analysis, which can be circumvented.
• Understanding information spread dynamics in social networks is crucial.

Purpose of the Study:

• To develop an unsupervised method for classifying misinformation in social media.
• To leverage user-user and user-media interactions for fake news detection.
• To model information propagation using network topology and entropy.

Main Methods:

• Created an experimental social media platform to simulate information spread.
• Developed a graphical model to analyze network topology evolution.
• Modeled uncertainty (entropy) propagation for fake and authentic media.
• Utilized user-user and user-media interaction data (e.g., likes).

Main Results:

• User-user and user-media interaction entropy approximates fake and authentic media likes.
• Demonstrated the ability to classify fake media in an unsupervised manner.
• Established a scalable approach to misinformation classification.

Conclusions:

• Network interaction patterns can effectively distinguish between fake and authentic information.
• Unsupervised learning based on network dynamics offers a promising alternative for misinformation detection.
• This approach bypasses the need for content analysis and ground truth labels.