Monitoring Social Tension Through Social Media Analytics

How can public institutions understand mood in the society in near real-time? We developed an AI-powered system that analyzes millions of Slovak-language social media posts to create a Social Tension Index — transforming unstructured online discourse into actionable intelligence for policymakers and researchers.

Challenge

Traditional methods of measuring public sentiment — surveys and polls — are slow, expensive, and capture only snapshots in time. Slovakia needed a way to continuously monitor mood in the society across diverse topics, from politics to public safety, using the vast amount of data people voluntarily share on social networks every day.

Beyond technical complexity, the challenge was linguistic and cultural. Slovak is a smaller language with limited NLP resources, and social media posts contain slang, sarcasm, emojis, and mixed sentiments that confuse standard algorithms. Building a system that truly understands Slovak online discourse required bridging AI capabilities with deep linguistic expertise.

Approach

We analyzed nearly 32 million Facebook posts and comments over one year using the XLM-RoBERTa-large multilingual model, fine-tuned specifically for Slovak. Through iterative training with real, synthetic, and multilingual datasets — and careful handling of edge cases like sarcasm and rhetorical questions — we achieved 87% classification accuracy across positive, negative, and neutral sentiments, categorized by 17 thematic areas.

The resulting Social Tension Index provides daily sentiment scores across topics like politics, crime, health, and conflict. It detects significant mood shifts — flagging when sentiment drops more than 25% below the rolling average — allowing institutions to correlate public emotion with real-world events and respond proactively rather than reactively.

Outcome

The system successfully captured major societal events: the Zochova street tragedy, the Tepláreň terrorist attack, and the anniversary of the Ukraine conflict all registered as significant sentiment shifts. With an average of nearly 60,000 posts analyzed daily, the index delivers continuous insight into Slovak public mood that no survey could match.

Beyond monitoring, the methodology opens new possibilities for official statistics on quality of life and social cohesion. It demonstrates how unconventional data sources — when analyzed responsibly and transparently — can complement traditional research methods, giving policymakers a more complete picture of the societies they serve.

Explore the full case

If you’d like to explore the full scope of this case study, read the publication with our contributions to experimental statistics using big data available for download (link: https://slovak.statistics.sk/wps/portal/ext/products/ssad/one/ssad-archiv/52148294-22a2-4356-8391-9a5e87ab8330/!ut/p/z1/rVNNU4MwEP01PabZkADhGB0FlFb5bJuLk9YqWIFamKL_3qDV8SKtM-aQ2Uzee5t9u8ESz7Gs1L54VG1RV-pZnxfSusuCW352RgRAPHHAz2KW3HDb8KYMzz4Aoe1_Avg0ugA_EbdudMUIMBNLfS3CMIyDLAM3My7Bp8SFaZoCRPaBf-4Kj9kBAA9cE3zhpZETUgqCfvC_899YiQmCZwLi4JrwkBz48MsScBp_4IEDfNenB_4AQA77l2GJ5apqt22OF_WyUTlqNiPg468YNY26H0G_I7Vb5cV-BKZBGDcchgxDGYhR00KcOgQ5ylxzWy25tq7X3a6Ke7w4CT071mg5bPOsz3ekk8c05OCwJNYBMGB23GzwQldi_1rJpYVn-2Ld4bSqd6We8PiPRnmAr7AsluW4W5VjGANl1AKLcM51SJht9Z_G2E3OJ49aWLU5KqqHGs9_dBDPT0qldYqnlxcp9ITUVbt-bfH8X0dkW6b9Kjl9Q5vI67rkIX9-euveAQepZmg!/dz/d5/L2dBISEvZ0FBIS9nQSEh/)

Download full version of case study as .pdf