Gravity of Policing: A Computational Framework for Measuring Force Intensity in News Coverage

Introduction: The portrayal of police force in news media shapes public perceptions of legitimacy and influences policy debates. Yet existing computational measures focus on sentiment or event categories, leaving a critical dimension unmeasured: the linguistic intensity of force with which police actions are described. This study introduces a "gravity of policing" framework—a continuous measure of how forcefully police action is represented in journalistic text, independent of crime seriousness or narrative stance.
Methods: We adapt valence–arousal–dominance (VAD) lexicons to construct a policing-specific gravity lexicon, weighting verbs and adjectives by dominance (power/control) and arousal (activation). A minimally supervised pipeline identifies police-focused sentences and computes article-level gravity profiles including mean, maximum, and proportion of tokens in low, medium, and high gravity bands. Eight benchmark incident types (petty arrest, protest policing, terror alert, terror incident, robbery, arson, attempted murder/murder, manslaughter) anchor interpretation, enabling the translation of raw scores into relatable narratives. Validation will employ human-coder ratings and large language model assessments on diverse English-language news corpora to establish convergent validity and distinguish gravity from generic sentiment.
Results: We present a measurement framework and algorithmic approach currently under development. We describe the lexicon construction process, feature extraction pipeline, and benchmarking strategy. Preliminary illustrations using example articles demonstrate how the method distinguishes between low-gravity ("police asked demonstrators to leave") and high-gravity ("police stormed the building and fired") portrayals of police action. Full validation results will be reported in subsequent work.
Conclusions: The gravity framework offers a theoretically grounded, transparent metric for policing-media analysis. It captures a force-intensity dimension orthogonal to sentiment, enables cross-context comparison, and provides an open-source tool for studying how police force is linguistically portrayed across diverse media settings. This advances computational social science by operationalizing a previously unmeasured framing dimension with broad applicability to research on policing, media effects, and institutional legitimacy.