ASGARD pursues a clear goal: strengthening the technological autonomy of law enforcement agencies (LEAs) and enabling their effective use of modern technologies. The focus is on forensics, intelligence, and prevention. Developed technologies are transferred to end users under an open-source approach, avoiding vendor lock-in. The project addresses the processing of seized data, big data solutions, and new research areas. A consortium of LEAs and research institutions ensures knowledge transfer and builds a closed open-source user community. In addition to traditional use cases and trials, hackathons are used to demonstrate results in practice. ASGARD follows an iterative approach with rapid results, emphasizing pragmatic solutions (“functional over perfect”), flexible deployment strategies, and a licensing/IPR concept tailored to LEA needs and ethical standards. Privacy, ethics, and societal impact are integral parts of the project. Thus, ASGARD creates a platform that allows investigators to benefit from modern technologies, agile methods, and open-source practices – the same approaches already used by the IT industry as well as organized crime and terrorist organizations.

Selected Results

Most of the research in this research project focuses on the development of new techniques, models and frameworks for the visual analysis of large amounts of social media interactions and text communications. As part of that we contributed:

A novel technique to support experts in their understanding of arbitrary, timestamped interactions, enabling a feature-driven investigation of relevant communication episodes, by modeling the bi-directional communication events, as a continuous communication density function. Additionally, we introduce features based on the communication density and other communication parameters which characterize the bi-directional communication behavior in individual episodes.

A visual analytics framework for the assessment of temporal hypergraph prediction models. We introduce its core components: a sliding window approach to prediction and an interactive visualization for partially fuzzy temporal hypergraphs. Our visual interface focuses on a single user in his context and combines detailed information on predictions, training, and holdout data.

A novel, interactive framework for temporal hypergraph exploration through the use of semantic zooming relying on a multi-level matrix-based approach and various exploration concepts. The technique incorporates a geometric deep learning model as a blueprint for problem-specific models while integrating visualizations for graph-based and category-based data with a novel combination of interactions for an effective user-driven exploration. We facilitate a focused analysis of relevant connections and groups based on interactive user-steering for filtering and search tasks, a dynamically modifiable partition hierarchy, various matrix reordering techniques, and interactive model feedback.

Apart of that the work also lead to advances in theoretical research, for instance, by shedding light on the similarities and differences of a multitude of projection techniques and the influence of features and parameters on data-representations and providing a data-driven intuition on the relation of projections. Postulating that depending on the task and data, a different choice of projection technique, or a combination of such, might lead to a more effective view.