Researchers report that analysing social media posts can offer early warnings of population movements, potentially helping humanitarian organisations respond more quickly. The study appears in EPJ Data Science and evaluates whether digital signals can complement or substitute traditional sources when surveys and field data are unavailable.
The research analysed nearly 2 million posts in three languages on X and used three case studies: Ukraine (10.6 million displaced), Sudan (approximately 12.8 million displaced) and Venezuela (about 7 million displaced). The team compared methods for classifying posts and found that sentiment labels (positive, negative, neutral) gave more reliable signals than discrete emotion labels (joy, anger, fear), especially for forecasting the timing and volume of cross‑border movements. Pretrained language models were the most effective early warning tools.
Marahrens explains the approach worked best in rapid conflict settings such as Ukraine and less well in slower economic crises like Venezuela. He cautions that social media analysis can trigger false alarms and is most valuable as an initial alert that prompts deeper investigation, particularly when combined with economic indicators and on‑the‑ground reports. The study also recommends exploring the relation between sentiment and emotion, using automated translation for more languages, and adding data from other networks to strengthen the tools.
The research received funding from the National Science Foundation and Georgetown University’s Massive Data Institute. Source: University of Notre Dame.
Difficult words
- analyse — examining information to find patterns or meaninganalysing
- population movement — people moving from one place to anotherpopulation movements
- humanitarian organisation — group that helps people in criseshumanitarian organisations
- complement — add to something to make it more complete
- pretrained language model — computer system trained on text before usePretrained language models
- sentiment label — short tag showing positive, negative or neutralsentiment labels
- false alarm — warning that suggests danger but is incorrectfalse alarms
- economic indicator — data that shows economic health or trendseconomic indicators
- automated translation — computer conversion of text between languages
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- How could humanitarian organisations use early warnings from social media before surveys arrive? Give reasons and examples.
- What are the main risks of relying mainly on social media analysis for population movement alerts?
- Which additional data sources mentioned in the article would you add to improve early warnings, and why?
Related articles
Glacial lakes and flood risk in the Hindu Kush‑Himalaya
The Hindu Kush‑Himalaya stores large freshwater in mountain glaciers. Warming has formed thousands of glacial lakes and raised the risk of sudden outburst floods; experts say better data sharing, observation and funding are needed but political and technical barriers remain.
China starts campaign to remove 'negative' online content
On September 22 China's Cyberspace Administration began a two-month campaign to remove online posts it calls antagonistic or negative. Platforms removed posts, banned influencers and targeted fan groups, trolls and so-called low-consumption vloggers.
Southern Trinidad villages face risks from U.S.–Venezuela tensions
Villages at Trinidad’s southern tip now face greater danger at sea as U.S. naval and air activity and Venezuelan warnings increase. Fishermen, migrants and local services feel the pressure; officials seek better safety and communication.