SocioFM Processed Societal Event Dataset, 2022–2025

This record contains the processed societal event streams used to train and evaluate SocioFM. The source records are public GDELT 2.0 event exports covering January 2022 through December 2025. The release contains 138,066,694 retained event records distributed across 48 monthly archives. Each shard includes filtered event-level sequences, observed country-day aggregates, serialized pretraining sequences, an acquisition and quality manifest, and a human-readable quality summary. Full news-article text is neither collected nor redistributed. Known unavailable source intervals and all filtering counts are documented in the supplied manifests and data card.