Mapping violence perceptions through YouTube comments: A new approach to real-time violence monitoring

We introduce a new dataset, the Violence Perception Index (VPI), which quantifies the intensity of references to violence in the online public discourse. Specifically, leveraging the YouTube API and natural language processing tools, the VPI captures public references to violence associated with 3,398,143 geolocated YouTube videos across Mexico (2020–2024). As such, it provides fine-grained temporal and spatial insights on violence perceptions, going beyond traditional event-based datasets. We benchmark the VPI against conventional conflict and violence indicators, demonstrating its validity and potential to enhance violence monitoring, early warning, and policy interventions. While the current dataset focuses on Mexico, the methodology is scalable across languages and geographies, making it adaptable to other conflict-affected settings. We discuss the methodological innovations and applications of the VPI, offering a significant advancement in understanding the diffusion and perception of violence through social media platforms.