AI applied to the analysis of extortion in Peru

Abstract

This study aims to digitally represent the extortion crisis in Peru between 2024 and 2025 through the analysis of YouTube comments, seeking to identify the most relevant discursive categories. A mixed-methods design with triangulation was applied, combining a quantitative component based on natural language processing techniques and a BERT model to classify comments, and a qualitative component that reconstructed a timeline of events and examined the relationship between peaks in digital conversation, transport strikes, and media coverage. Findings reveal that digital conversation is not triggered by isolated crimes but by strikes that compel mainstream media to cover the issue, producing abrupt increases in online interaction. The most frequent narratives focus on blaming the state and demanding harsher punitive measures, reflecting a citizenry that is both indignant and supportive of stricter security responses. Media coverage appears delayed and reactive, postponing the inclusion of the issue in the public agenda and prolonging the perception of vulnerability. This work provides a replicable framework for studying criminal phenomena in digital environments and highlights the role of social protest as a catalyst for online visibility and debate. It offers valuable insights into how citizen pressure can shape media and political agendas, underscoring the importance of integrating digital discourse analysis into the design of public security policies.