Multimodal graph analysis for law enforcement

This project proposal addresses the central challenge of modern digital forensics, which is characterized by increasing fragmentation and vast quantities of heterogeneous data sets. In current investigative practice, criminally relevant digital traces—such as texts, images, audio, and video files—are mostly viewed in isolation and in a modality-specific manner, which complicates the systematic linking of distributed evidence and carries the risk of overlooking complex correlations. The project tackles this deficit by aiming to conceptually overcome this methodical isolation through the use of modern, multimodal AI methods. The core of the research approach lies in transforming heterogeneous content into a unified, spatial, and temporal graph model that not only maps relationships between people, objects, and events but also makes them analytically accessible. A significant focus is placed on the explainability and traceability of the methods used to meet the high demands of the judicial environment regarding documentation and legal admissibility. The project explicitly views itself as a scientifically grounded examination of the methodological foundations of forensics rather than the mere development of an operational tool. By systematically investigating explainable, graph-based approaches, the project makes a significant contribution to the further development of digital investigation strategies. It establishes the basis for capturing large and complex evidence more efficiently, performing context-based prioritization, and reconstructing plausible sequences of events along a consistent timeline, which is of considerable methodological interest, particularly for work within existing legal frameworks.