Label propagation in a distributed system

A computer-implemented method when executed by one or more processors causes the one or more processors to perform operations comprising:

• obtaining, during an iteration of a label propagation algorithm, a set of vertices from a graph representing relationships among data items;
• obtaining a first label value indicting a strength of association between a first label and a first vertex of the set of vertices;
• receiving an incoming message from a first computing system via a network, the incoming message comprising a weighted source label value for a source vertex having one or more outgoing edges to the first vertex, the weighted source label value computed by the first computing system;
• in response to receiving the weighted source label value, determining an updated first label value for the first vertex based on the first label value and the received weighted source label value;
• determining that the updated first label value exceeds a threshold value;
• in response to determining that the updated first label value exceeds the threshold value, assigning the first label to the first vertex;
• updating one or more outgoing edges of the first vertex using the updated first label value;
• and
• updating, using the one or more outgoing edges, by sending an outgoing message to a second computing system via the network, a second vertex of the set of vertices that neighbors the first vertex, the outgoing message comprising a weighted first label value for the first vertex, the second computing system to update a second label value for a second label of the second vertex based on the weighted first label value.