In plain English, this service looks at which websites link to a particular target website, and then it ranks websites that are popular among those linking websites using a method commonly used in recommendation algorithms.

In technical jargon, it reinterprets the incident edges in the adjacency matrix as sparse high dimensional vector, and uses cosine similarity to find the nearest neighbors nodes within this feature-space.

This is a write-up about an experiment from a few months ago, in how to find websites that are similar to each other. Website similarity is useful for many things, including discovering new websites to crawl, as well as suggesting similar websites in the Marginalia Search random exploration mode.