Akademik Veri Yönetim Sistemi
Spectrum of Engineering and Management Sciences, cilt.4, sa.1, ss.55-76, 2026 (Scopus)
Online platforms have emerged as essential infrastructure for democratic deliberation, crisis communication, and public health messaging. Nonetheless, governance of these spaces remains largely retrospective and principle-based, without concrete evaluative criteria. This article operationalizes six key principle-based elements of platform governance into concrete clause components with quantifiable triggers, to support proactive application. We develop a composite Governance Risk Index (GRI), integrating empirically defined decision thresholds for four risk components: dispersion, drift, inequality, and toxicity. We estimate toxicity levels from a stratified subsample of the dataset labeled using the Perspective API, aggregated to the daily level. Our results show that toxicity is positively correlated with engagement (r = 0.52, p < .001), as replicated in our data (r = 0.49, p < .001), in line with algorithmic amplification dynamics that platform governance must account for. We further find that decision thresholds differ meaningfully between clause components and risk components. A contextual benchmark against 2015–2016 multi-platform baselines reveals that the 2024 Twitter/X environment exhibits substantially different statistical properties in engagement dispersion and toxicity prevalence, highlighting the need for context-dependent calibration of platform governance thresholds. Over a 30-day out-of-sample holdout period, validated exclusively against independently verified external events, the GRI classified governance-event days under retrospective validation with 85.1% accuracy. Under a supplementary labeling scheme that incorporates platform-internal anomaly criteria, accuracy reaches 90.0%, representing an improvement of about 13 percentage points over single-metric baselines, highlighting the advantages of multivariate, multi-faceted operationalization in proactive platform governance.