Chinese Whispers

Constellation Action	Keyboard Shortcut	User Action	Menu Icon
Run Chinese Whispers		Tools -> Cluster -> Chinese Whispers

Chinese Whispers is a clustering algorithm which forms clusters around the strongest classes in the graph.

The algorithm works as follows:

1. Firstly it assigns all nodes in the graph their own class
2. It then iterates through each node in a random order. For each node, it changes its class to the one which has the largest sum of edge weights connected to that node. If there is a tie for the largest, it randomly chooses one of the largest.
3. Step 2 is repeated until clusters stabilise.

Because of the random nature of the algorithm, multiple runs of the algorithm on the same graph can produce different clusters (depending on the order nodes are iterated through, and which classes are selected in the case of ties).

Constellation Display

Chinese Whispers in Constellation makes use of overlay colors. When the clustering algorithm has been run, each cluster is assigned a unique color. Node backgrounds and intra-cluster transactions are colored using the cluster's color, while inter-cluster transactions are colored dark grey.

Rather than set the color directly, Chinese Whispers in Constellation creates a new color attribute "Cluster.ChineseWhispers.Color" and tells Constellation to display the graph using the new attribute (via the "node_color_reference" and "transaction_color_reference" graph attributes). To switch back to the default color attribute, edit the graph attributes via the Attribute Editor and click on "Restore Default" followed by Ok.

Before Chinese Whispers is run:

After Chinese Whispers is run:

When the clustering algorithm runs, no arrangement is done, so the graph can look confusing.

The clustering algorithms add an integer attribute "Cluster.ChineseWhispers" to the nodes. This attribute has a unique per-cluster value to indicate which cluster a node belongs to, which other algorithms can later use (e.g. Arrange by Node Attribute).