Work

Geo-social media beyond Twitter: Comparing retrieval strategies for Reddit, TikTok, Telegram, Mastodon, and Bluesky during the 2024 Central European floods

Year: 2026
Type: article
Abstract: Social media can provide timely information during floods, but cross-platform keyword retrieval remains challenging because search interfaces and access conditions differ. This study retrospectively compared English- and German-language posts retrieved from Bluesky, Mastodon, Reddit, TikTok and Telegram during the September 2024 Central European floods. Two strategies - phrase-based queries (Test 1) and root-based, stem-like queries (Test 2) - were applied using equivalent keyword sets adapted to each platform’s constraints. Performance was evaluated in terms of retrieval volume, duplication, disaster relatedness, relevance and geoparsed location references. Spatial patterns were analysed through H3 aggregation and spatial autocorrelation and validated against a satellite-derived flood mask. Bluesky and TikTok yielded the largest volumes (11,232 versus 49,386 posts and 9718 versus 25,560 posts, respectively, for Tests 1 and 2), whereas Reddit (246 versus 249) and Telegram (85 versus 83) had substantially fewer posts. Root-based queries substantially increased retrieval volume on Bluesky, Mastodon and TikTok, but phrase-based queries produced a higher proportion of relevant content. About one third of geocoded related-and-relevant posts (33.1% and 34.4%) were geoparsed to the United States, showing that global retrieval captures international discourse alongside local observations; spatial analyses were therefore restricted to Central Europe. At H3 Level 4, 83.81% of occupied cells under Test 1 and 81.82% under Test 2 intersected the flood mask, exceeding the 67.73% spatial baseline. Cells containing posts from multiple platforms represented 40.00% and 36.36% of the respective spatial footprints. Multi-platform integration increased overall spatial completeness and reduced the likelihood of platform-specific blind spots.
Language: English
Open Access status: hybrid
Estimated APC paid: $3,200
OpenAlex ID: W7226033919
FWCI:
Cites: 44
Cited by:
Related to: 10