SEVAL
Whether in connection with real or virtual events, the digital space now acts as a central medium in which information is rapidly provided and disseminated due to global networking and mobile devices. This resource offers great, largely untapped potential for civil protection in Germany - from the early detection of individual events in public spaces, such as technical disruptions, to the detection of multiple events whose effects can influence each other, to the holistic consideration of physical and digital developments that may be interrelated.
This potential faces a variety of challenges with regard to active information extraction from web data for control centres and situation centres, such as distributed and large amounts of data to be processed with limited resources. The transfer of approaches already known in research into practice usually requires expert knowledge. In addition, the use of information for decision-making touches on issues of data confidentiality.
In SEVAL, the practical development of web data as a reliable and operationally manageable source of information for the complementary expansion of the situation picture is therefore being researched and demonstrated. In particular, the robust identification of potentially relevant information in various online sources, the geolocalisation of text data and the contextualisation and evaluation of information for civil protection tasks are addressed.
As a contribution to the BBK's all-hazards approach, selected events and event types are used to investigate the extent to which current approaches to text analysis (Natural Language Processing, NLP) can provide practicable collection, analysis and provision of web data as complementary information for situational awareness. The central question here is whether and how events relevant to civil defence can be identified more quickly and comprehensively. The focus here is on aspects of simple applicability and transferability of the methods used to different sources and event types. The focus is on current AI models and their application for filtering, geolocalisation and linking of information as a basis for plausibility checks and evaluation of text data and images.
In addition to analysing the achievable accuracy of the models, the possibilities and limitations with regard to the new data sources and methods addressed here for practical use in civil protection are examined and highlighted.
The project is funded by the Federal Office of Civil Protection and Disaster Assistance.
Project duration: 10/2024 - 09/2027