Labeled Disaster Location Descriptions from Twitter/X for Ten U.S. Events

United States disaster data containing labeled location descriptions from social media posts during ten events, including hurricanes, floods, wildfires, tornados, and winter storms. The dataset was created by KAI SUN and published on figshare, with a last update in April 2026. It is intended for studying how people describe locations in disaster contexts and for training AI models to extract them.

Use Cases

• Training named entity recognition models to extract detailed location mentions like addresses and intersections from disaster-related text.
• Studying linguistic patterns and multi-entity location descriptions used in crisis communication on social media.
• Developing AI tools for automated situation awareness by parsing victim, damage, and resource locations from disaster tweets.

Strengths

• Focuses on ten distinct disaster events across five disaster types, providing contextual variety.
• Explicitly contains detailed, multi-entity location descriptions which are noted as limited in other datasets.
• Messages are sourced from the social media platform Twitter/X, reflecting real-world communication patterns.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale model training.
• Data may reflect geographic and platform bias inherent to its collection from Twitter/X for U.S. disasters.

Provenance

Source

Twitter/X social media platform.

Collection Method

Collection, labeling, and validation process described in the associated paper.

Freshness

Last updated 2026-04-23 04:40:08; freshness should be verified.

Geography

United States.