A Few Hundred Accounts Are Quietly Steering A Nation Apart

A surprisingly small circle of social media users can reshape an entire country’s political divide, and Finland offers a vivid case study. In a new paper in the journal Network Science, researchers from Aalto University use Twitter data from the 2019 and 2023 Finnish parliamentary elections to show that elite accounts contribute far more to online polarization than their numbers would suggest. Their method dissects polarization into separate layers for elites and ordinary users, revealing who is actually driving the split and how that influence has changed over time.

At the heart of the work is a simple discomforting idea. When we talk about a polarized public, we often imagine millions of people drifting into opposing camps. But this study suggests that much of the structural divide in online debates around climate, immigration, the economy, social policy, and education can be traced to a dense core of highly connected accounts. Around them, a far larger “mass” of users mostly amplifies what the core is already doing.

Pulling The Network Apart, Layer By Layer

The authors start from a now standard move in computational politics: represent political discussion as a network. Twitter users become nodes, and retweets form links interpreted as endorsements or shared viewpoints. For each election year, they assemble five separate networks, each centered on a major topic such as climate or immigration, using 12 weeks of tweets leading up to election day.

Conventional polarization metrics typically compress that complexity into a single score that captures how cleanly a network splits into two camps. The Aalto team builds on this tradition but pushes further. They use a stochastic block model to detect two polarized groups in each topic network, then apply a core periphery framework inside each group. The densely connected “cores” are treated as elites, and the sparser “peripheries” as masses.

Crucially, the authors do not assume in advance who counts as elite. Status is inferred from the network structure itself, through how central and densely connected an account is. That choice allows the method to capture politicians, media figures, activists, and other opinion leaders on the same footing, as long as they occupy the structural core.

With elites and masses identified, the team decomposes a leading polarization score, the Adaptive EI-index, into distinct components. They track elite cohesion (how tightly elites in a camp connect to one another), mass amplification (how strongly the masses retweet their own elites), mass cohesion, and the frequency of cross camp “bridge” ties that cut against polarization.

Elites Are Few, But Their Polarizing Power Is Large

Once the decomposition is in place, some stark asymmetries emerge. Across all issue networks and both election years, elites are consistently more responsible for structural polarization than their modest head counts would suggest. In immigration debates, for example, right leaning elites made up less than 3 percent of users in 2019 yet were responsible for over 30 percent of the observed polarization. By 2023 they were still only about 4 percent of users, but they contributed roughly 20 percent of the structural divide. In education discussions in 2023, right leaning elites represented around 2 percent of accounts and about 18 percent of polarization.

The climate network shows how these patterns can shift over time. In 2019, the left leaning camp accounted for about 72 percent of polarization on climate, dominating the structure of the debate. By 2023, that dominance had faded to about 25 percent. Right leaning elites became far more consolidated, and their masses showed big jumps in both amplification and cohesion, meaning more of their periphery was endorsing and connecting within the camp.

The authors are careful to note that group size and sheer activity do not reliably predict these contributions. In some topics, the more polarized side is actually the smaller one. Activity spikes near the elections do not automatically translate into higher polarization either; what matters is how that activity is organized inside the network’s hierarchy.

The Rise Of Alignment, And The Road To Gridlock

Beyond structure, the study also examines issue alignment, the degree to which a person’s stance on one topic predicts their stance on another. Using information theoretic measures, the authors compute alignment separately for elites and masses across pairs of topics.

Elites are more aligned than the masses in every topic pair and in both years, and they became even more aligned by 2023. In most issue pairs, elite alignment reached very high values, indicating that knowing where an elite stands on, say, climate tells you a great deal about how they talk about immigration, social policy, or the economy. The masses started from much lower alignment in 2019 but experienced a sharper proportional increase, with their mean alignment rising from about 0.13 to 0.44.

One quote from the Aalto team captures what that means for democracy:

“In democracies, it’s healthy, even desirable, to disagree sharply on individual issues. But when alignment becomes complete, society splits into just two camps that disagree on absolutely everything, and there’s nothing left in common with the other side,” says Mikko Kivela, Professor of Computer Science at Aalto University and co author of the study.

The worry is not disagreement itself, which is normal and often productive. It is the locking together of positions across many issues, so that every topic becomes another front in the same underlying conflict. The paper suggests that in Finland’s Twitter based political sphere, elites are already close to that kind of complete alignment, and the masses are moving in the same direction.

The study also shows that cross group bridge ties, which are the only interactions that mechanically reduce the polarization score, have generally declined or stayed flat. In climate and social networks in particular, the share of interactions that cross the aisle drops over time, underscoring how hard it is for bridge building to keep up with intensifying internal cohesion.

As lead author Ali Salloum puts it, the consequences can be severe:

“One of the most serious consequences of polarisation and alignment, beyond the threat of political violence, is political gridlock. Legislation slows down and weakens, poor quality decisions are made, or no decisions are made at all,” says Salloum.

Beyond Finland, And Behind The Data Wall

Although the paper focuses on Finnish elections, the authors explicitly design their method to be portable. Any setting where political conversation leaves a traceable network, from other countries’ election campaigns to issue specific movements, could be analyzed using the same decomposition. That opens the door to more precise diagnoses of whether a polarized environment is being driven mainly by elite cohesion, mass mobilization, or some combination.

But the study arrives at an awkward moment for researchers. The authors note that Twitter, now X, has sharply restricted access to programmatic data, and future work may have to move to newer platforms such as Bluesky. The irony is hard to miss: just as methods for reading the structure of polarization become more sophisticated, the underlying data streams are being closed off.

The core message, though, is clear. Polarization is not simply a story about two generic sides drifting apart. It is about who sits at the center of each side, how tightly those cores are wired together, how strongly the masses fall in line, and how rarely people still talk across the divide. By turning those intuitions into measurable components, the Aalto team offers both a sharper warning and a more precise tool for understanding how a small number of elite accounts can steer a nation’s arguments, and perhaps its politics, onto more rigid and fragile paths.

Study: “Anatomy of elite and mass polarization in social networks,” Network Science, Volume 13, 2025, e18. DOI: 10.1017/nws.2025.10010.