A New Computational Method for Quantification and Analysis of Media Bias in Cybersecurity Reporting

The proliferation of cyber threats and their representation in news media significantly influence public awareness and policy-making. However, biases in cybersecurity news reporting—manifested as location focus, event significance prioritization, and coverage of specific attack types or industries—distort perceptions and resource allocations. This study addresses these challenges by developing an artificial intelligence-driven computational framework that analyzes and quantifies biases in cybersecurity news. Leveraging a dataset of 1.23 million news articles, which was filtered down to 9314 cybersecurity-related events spanning 144 sources over five quarters, the methodology employs GPT-based classification, Shannon entropy, chi-square tests, multinomial logistic regression, and Bayesian inference to identify patterns and dependencies in reporting. Results reveal that generalized sources such as BBC exhibit high reporting diversity (H = 2.87), while specialized outlets such as Cybersecurity Insider display niche focus (H = 0.45). Multinomial logistic regression achieved 81.2% accuracy in predicting reporting tendencies based on event significance and source characteristics. Bayesian inference highlighted significant tendencies, such as Dark Reading’s preference for advanced persistent threats (posterior probability = 0.62) and CNBC’s emphasis on phishing events (posterior probability = 0.48). Furthermore, this framework can be extended to other open-source media, such as social media platforms, to pinpoint and measure the level of bias exhibited by individual users, marking a significant step toward identifying misinformation, disinformation, and propaganda. The proposed framework not only provides actionable insights for stakeholders but also establishes a scalable model for analyzing biases in other domains, fostering informed decision-making and balanced reporting practices.