How to Read These Graphs?

What Do These Graphs Show?

These graphs are interaction maps: they show which social media accounts interacted with each other’s posts during the monitored period. Each dot represents one account, and each line between two dots indicates that those accounts interacted — one account responded to the other’s content.

Think of it as a map of conversations: who reacted to whose content, and how do these accounts form groups?

During the monitored period, between 6,824 and 6,826 accounts were active in the dataset. For readability, each graph only displays the 87–89 most connected accounts, roughly the top 1–2% based on engagement levels. The remaining thousands of less-active accounts were filtered out to prevent the graph from becoming an unreadable mass of dots.

Size of the Dot = Level of Engagement

The larger the dot, the more interactions that account had during the period. The size is based on the total number of unique accounts that either engaged with the account or that the account actively responded to.

In this network, the largest dots are accounts such as telegraaf.nl, DENK, NOS Stories, Forum voor Democratie, and nu.nl — all of which attracted hundreds of unique responders during the six-week period. The smallest visible dots only had a handful of interactions.

The dot size has a maximum cap — the very largest dots all share the same maximum size, even if their interaction numbers differ by hundreds. This is intentional: beyond a certain point, the exact number matters less than the fact that the account functions as a dominant hub. Hover over a dot to see the exact figures.

Shape of the Dot = Type of Account

A regular social media account. The majority of accounts in this network are circles.
A followed politician — an account in the database with a known party affiliation (such as DENK, FvD, VVD, etc.). Diamond-shaped nodes also appear with the ◆ symbol in the top account table. A politician node embedded in a cluster of accounts with high bot scores may indicate coordinated activity directed at or reinforcing that politician.

Thickness of the Line = Frequency of Interaction

The thicker the line between two accounts, the more often those accounts interacted during the monitored period. A very thin line means they interacted only once. A thick line means there were many back-and-forth interactions, indicating a strong or repeated engagement relationship.

Minimal Interaction
Moderate Interaction Frequency
Frequent interaction and strong engagement

Direction of the Arrow (Directed Graph Only)

In the directed graph, each line contains a small arrowhead. The arrow points from the content creator toward the responder.

If a line points from Account A to Account B, this means that Account B responded to posts made by Account A.

When hovering over a line, a small label appears showing the direction and number of interactions, for example:

→: 3 interaction(s)

What to Look For in the Connections

Multiple Thick Lines from One Dot: An account with multiple thick lines repeatedly interacts with several other accounts. For content creators (high out-degree), this means their content consistently attracts the same responders. For responders (high in-degree), this means they repeatedly return to content from the same accounts.
Thin Lines Between Clusters: In the undirected graph, if you see a single thin line connecting two otherwise separate color clusters, this represents the only bridge between those two communities. If that one connection disappeared, the two groups would become completely isolated from each other.
Reciprocal Connections: In the directed graph, if arrows exist in both directions between two accounts (A→B and B→A), this means the accounts mutually respond to each other’s content. This reflects a genuine two-way conversation, which is less typical for automated behavior.

What is a Cluster?

A cluster (also called a community) is a group of accounts that interact more with each other than with accounts outside the group. Clusters are automatically detected using an algorithm called Louvain. The algorithm has no knowledge of the content, political orientation, or identity of the accounts. It only analyzes interaction patterns. You can think of clusters as online conversation bubbles: groups of accounts operating within their own corner of the interaction network.

How to Interpret Cluster Membership

If two accounts appear in the same cluster, this does not necessarily mean they agree with each other — it simply means they interacted. For example, a journalist and a politician being criticized by that journalist could still end up in the same cluster if they regularly interacted with each other’s content.

Modularity Score: How Strong Is the Clustering?

Each cluster receives a modularity score between 0 and 1. A higher score means the cluster is more self-contained — its members interact more with each other than with outsiders. A score above 0.60 is generally considered strong clustering. All six clusters in this network score between 0.60 and 0.74, indicating that the clustering is genuine and not simply an artifact of the algorithm — these accounts predominantly interact within their own groups.

Why This Matters

High Modularity = Information Silos: When clusters are strongly self-contained, narratives circulating within one cluster do not naturally spread to other clusters. Bridges between clusters are the only pathways through which information can move between groups.
Concentration of Bots Within a Cluster: If a cluster contains many accounts with orange or red borders, the interaction patterns of the entire cluster may be artificially inflated.
Small, Highly Connected Clusters: An unusually small cluster (such as Cluster 5 containing only four accounts) with high modularity means those four accounts interact almost exclusively with each other. This may indicate coordinated activity.

Municipal Elections 2026: Part II