Page 202 - AI for Good Innovate for Impact
P. 202
AI for Good Innovate for Impact
Use Case 5: LLM-driven creation of natural hazard geodatabase
from digital mass media
Organization: HSE University
Division: Faculty of Geography and Geoinformation Technology
Country: Russian Federation
Contact Person(s):
Anna Derkacheva (aderkacheva@ hse .ru)
Maria Sakirkina
Gleb Kraev
Tatiana Aniskina
1 Use Case Summary Table
Item Details
Category Climate Change/Natural Disaster
Problem Addressed Traditional natural hazard (NH) databases are fragmented, manually
maintained, and often inaccessible, limiting real-time situational aware-
ness and policy planning.
Key Aspects of Solu- - LLM-based automated extraction from mass media
tion - Geocoding and integration with GIS
- Capture of local and low-scale events
- Quality estimation and verification pipeline
Technology Keywords LLM, Natural Hazard Monitoring, Geodatabase, Text Mining, GIS, Disas-
ter Risk Management
Data Availability Yes – demo datasets available online (in Russian)
Metadata (Type of Natural hazard type, location, date/time, socio-economic impacts, emer-
Data) gency response, event intensity
Model Training and Custom prompts and LLM pipelines used for extracting structured data
Fine-Tuning from unstructured Russian-language digital news; additional automated/
manual QA filters applied
Testbeds or Pilot Phase 1 processed 8 million articles – 52 000 NH-relevant texts – 38 000
Deployments events (Russia, 2018–2024)
Repository URLs Demo Dataset & Background (in Russian)
166
