Page 203 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

2 Use Case Description

2�1 Description

Building a complete picture of natural hazard (NH) exposure requires extensive, standardized, Change 4.2-Climate
and frequently updated databases. However, in Russia and elsewhere, NH data is fragmented
across governmental and institutional silos, often manually updated and inaccessible for public
or academic use.

This use case addresses that gap by using a generative Large Language Model (LLM) to
mine natural hazard reports from open-access digital mass media. These are transformed into
structured geospatial datasets, suitable for integration into GIS and disaster risk frameworks.

Approach Highlights:

• Massive-scale textual input (8M publications) filtered and processed using LLMs
• Events geolocated using GIS-compatible structures
• Captures socio-economic damage, response efforts, and event severity
• Includes built-in data quality controls and transparency reports

The system enables national and local event capture, empowering risk analysis and urban
resilience planning. It serves as both a research tool and practical risk database for future
disaster mitigation.

Use Case Status: Part of ongoing product development with operational dataset and pipeline
running

Partners

• No external partner organizations reported at this time. All development and deployment
are led by the internal team at HSE University.

2�2 Benefits of use case

• Strengthen resilience and adaptive capacity to climate-related hazards and natural
disasters.
• Reduce the number of people affected and the economic losses caused by disasters.

2�3 Future Work

• Archival Data Expansion: Add earlier historical records from news archives to enrich the
timeline.
• Quality Enhancement: Improve accuracy of input filtering and LLM output validation.
• Geocoding Improvements: Develop more precise location matching and error reduction
tools.
• Usability Extension: Build full-featured dashboard with visual mapping, time-series tools,
and data export.
• Internationalization: Extend pipeline to other languages and countries beyond Russia.
Integration: Connect database with emergency preparedness agencies and city planning tools.

167