Ghost Authors and Hidden Sources: Structural Metadata as a Forensic Indicator of Global Online Disinformation

The surge of internet misinformation has generated increasing interest in automated detection methods that can function on a large scale. NLP techniques have predominantly influenced the literature, although structural metadata indicators like author attribution and source provenance remain inadequately examined as forensic signals. This research introduces a digital forensics methodology utilised on a categorised collection of 580 online news items sourced from the Information Security and Object Technology (ISOT) dataset and the FakeNewsNet (FNN) corpus. The system functions through three consecutive stages: structural metadata analysis, language characterisation, and automated validation. Phase 1 results indicate that worldwide disinformation exhibits total structural anonymity, lacking both author identification and source transparency at 0%, in contrast to the 100% attribution seen in authentic global media. Statistical verification by Fisher’s Exact test established markedly significant correlations (p < 0.001). Phase 2 linguistic examination revealed quantifiable distinctions in readability, subjectivity, and sentiment polarity, with Global Real news exhibiting the greatest textual intricacy and the most impartial emotional disposition. The Phase 3 machine learning evaluation demonstrated that the incorporation of structural metadata and linguistic attributes with TF-IDF text representations enhanced classification efficacy across multiple classifiers, with Gradient Boosting attaining the peak accuracy (87.1%) and F1-score (86.7%) in the combined setup. McNemar’s tests suggest that performance improvements were consistently directional, however not statistically significant at the 5% threshold, a conclusion attributed to limitations in sample size. The research provides empirical evidence for the forensic value of structural metadata and suggests a hybrid feature approach for implementing misinformation detection in international news contexts.