Real social-graph clustering doesn't insist each cluster be a perfect clique — it asks for dense subgraphs that nearly cover the vertex set. The Louvain method, label propagation, and Leiden algorithm (used by LinkedIn, Facebook, Twitter for community detection) are clique-cover relatives that optimize modularity rather than exact cover count. Same ancestry: minimize the number of groups while keeping each group internally well-connected. The exact clique cover problem itself shows up in data compression — a clique in a feature co-occurrence graph is a set of always-together features, mergeable into one — and in chip layout, where clique cover the standard-cell library helps macro selection.
39
Community detection
When LinkedIn's "people you may know" partitions the social graph into clusters, it is doing a relaxed clique cover.