Network Visualization of Criminal Co-offending Patterns Using GPT-4: Analysis of Social Connections in Organized Crime

Abstract

Crime continues to pose a significant threat to community stability and socioeconomic development in our global society. Artificial intelligence technologies are increasingly being deployed to predict potential criminal activity and analyze the structural patterns within criminal networks through data-driven approaches. This paper introduces a novel methodology utilizing GPT-4 tools to examine social connections within criminal networks through graph-based visualization techniques. We developed a graph visualization approach for criminal data that effectively identifies structural patterns within criminal organizations. Our research analyzed 2,113 criminal cases related to vehicle theft, robberies, and armed robberies between 2013 and 2024 in the Ternopil region, resulting in visual network models of criminal co- offenders. By leveraging GPT-4's multimodal capabilities, we processed criminal data and generated graph representations illuminating the social connection structures among offenders. Our findings reveal distinctive network patterns across different crime types: vehicle theft networks demonstrate complex, highly centralized structures with key coordinator roles; robbery networks typically feature small, stable groups of 2-3 individuals, reflecting the operational requirements of such crimes; and armed robbery networks exhibit larger (4-6 person), more structured organizations with clear role distribution, likely due to the need for violence coordination and victim control. This methodology offers law enforcement agencies an effective analytical tool for addressing organized crime in contemporary settings.