How Social Media Sentiment Is Emerging as a Powerful Early Warning Signal for Human Displacement During Crises

A growing global crisis is unfolding as forced displacement continues to rise at an alarming pace.

Over the past decade, the number of people uprooted from their homes has nearly doubled, according to the United Nations’ refugee agency. In 2024 alone, one in every 67 people worldwide was forced to flee, illustrating just how widespread and urgent the issue has become. Against this backdrop, researchers are exploring new ways to anticipate when populations are likely to move during wars, civil conflicts, and economic collapses. One promising approach, highlighted in a recent study led by University of Notre Dame researcher Helge-Johannes Marahrens, shows how analyzing social media sentiment might help predict these movements more effectively—potentially allowing humanitarian organizations to prepare sooner and respond faster.

This new work, published in EPJ Data Science, digs deep into how people express themselves online, especially in moments of uncertainty or danger. Traditionally, early-warning systems have relied heavily on surveys, official reports, and on-the-ground assessments. But in forced migration scenarios, these tools often break down. It’s nearly impossible to collect reliable data when people are fleeing rapidly or when communication networks are disrupted. The study argues that digital traces—like social media posts—can fill in crucial gaps and offer real-time signals that traditional methods simply cannot provide.

What the Researchers Examined

To understand how online expression correlates with physical movement, the research team examined nearly 2 million posts from the platform X (formerly Twitter). These posts were written in three different languages, corresponding to three major displacement crises:

Ukraine, where 10.6 million people were displaced following Russia’s full-scale invasion in 2022
Sudan, where a civil war that began in April 2023 displaced roughly 12.8 million people
Venezuela, where ongoing economic and political turmoil over recent years has pushed approximately 7 million people to leave their homes

The researchers focused on two main aspects of language in the posts:

Sentiment — whether the tone of the message was positive, negative, or neutral
Emotion — specific emotional categories such as joy, anger, or fear

They tested multiple computational approaches to categorize posts, ranging from simple rule-based tools to advanced pretrained language models that learn patterns from massive amounts of text. These models, powered by deep learning, are designed to interpret language in a way that’s more flexible and context-aware—similar to how humans understand nuance.

What They Found About Predicting Movement

One of the clearest conclusions from the study is that sentiment is significantly more useful than emotion for predicting forced displacement. Negative sentiment, in particular, tended to rise shortly before large movements of people occurred.

In crisis environments such as Ukraine’s, sentiment signals were strongest. The study found that shifts in negative sentiment often preceded surges in cross-border migration, making it a potentially valuable early warning indicator. Emotion categories like fear or anger, on the other hand, were less consistent and appeared less frequently in the posts. In some cases, emotions were so sparse that they carried little predictive value.

The type of crisis also mattered. Armed conflicts and sudden outbreaks of violence produced sharp, noticeable changes in online sentiment that correlated better with physical movement. But in slower-developing crises—like Venezuela’s long-term economic decline—the sentiment signals were weaker and less tied to migration timing.

Across the board, pretrained language models outperformed older analysis techniques. These models handled multilingual data more effectively and were better at detecting subtle shifts in sentiment, especially in noisy or ambiguous posts.

Why This Matters for Humanitarian Response

The potential use of social media as an early-warning system is promising for several reasons. First, digital data is produced constantly and at scale, whereas surveys and field assessments take time and may arrive too late during fast-moving crises. If humanitarian organizations can monitor sentiment shifts and detect early signs of instability or fear, they could:

Pre-position supplies
Deploy teams earlier
Prepare border infrastructure
Coordinate with local governments and NGOs
Launch targeted communication campaigns

Even a few days’ notice can improve outcomes for displaced communities by allowing for more organized support.

The study does warn, however, that sentiment analysis should not be used in isolation. Social media posts represent only a portion of the population, and many of the most vulnerable individuals may lack internet access or avoid public platforms. Sudden spikes in negative sentiment can also produce false alarms if they are unrelated to migration triggers. Because of this, the researchers recommend using sentiment signals as an initial flag—something that prompts a deeper investigation involving traditional data sources.

The Study’s Limitations

While the findings are compelling, the authors emphasize several limitations:

Not everyone is active on social media, leading to potential biases
Language coverage is uneven, making it harder to analyze posts from regions with fewer digital resources
Emotion detection remains unreliable due to its complexity and cultural variance
Digital expressions do not always translate directly into physical behavior
Social media signals can be manipulated, either unintentionally or intentionally

These constraints mean that sentiment analysis is best viewed as one piece of a broader early-warning toolkit, rather than a stand-alone solution.

Additional Insights About Forced Displacement Trends

Since this topic goes far beyond a single study, it’s helpful to consider broader context as well. Forced displacement usually falls into three major categories:

Conflict-driven displacement — caused by war, violence, persecution, or political instability
Disaster-related displacement — triggered by natural disasters, climate events, or ecological disruptions
Economic displacement — driven by the collapse of economic conditions, hyperinflation, unemployment, or shortages

Each type produces different migration patterns. For instance:

Conflicts produce sharp spikes in movement, which aligns well with sentiment-based predictions
Economic crises often lead to gradual, long-term migration, making early detection more challenging
Climate-related displacement is becoming more frequent and may combine fast-onset events (like hurricanes) and slow-onset pressures (like drought)

Digital tools like sentiment analysis could eventually become part of global humanitarian monitoring systems that track migration risk across multiple countries simultaneously.

What Future Research Might Explore

The authors suggest several promising directions for future study:

Exploring how sentiment and emotion intersect and whether combining them improves predictions
Using automated translation to analyze more languages and reduce data gaps
Including other social platforms, not just X, to increase coverage
Linking sentiment signals with traditional indicators such as economic models, conflict forecasts, and field reports

With every improvement in data quality and model sophistication, organizations gain a clearer picture of how and when populations are likely to move.