Mining Contrasting Quasi-Clique Patterns
Authors: Roberto Alonso, Stephan Günnemann
Summary: Mining dense quasi-cliques is a well known clustering job with functions starting from social networks over collaboration graphs to doc evaluation. Current work has prolonged this job to a number of graphs; i.e. the aim is to search out teams of vertices extremely dense amongst a number of graphs. On this paper, we argue that in a multi-graph situation the sparsity is effective for information extraction as nicely. We introduce the idea of contrasting quasi-clique patterns: a group of vertices extremely dense in a single graph however extremely sparse (i.e. much less linked) in a second graph. Thus, these patterns particularly spotlight the distinction/distinction between the thought of graphs. Based mostly on our novel mannequin, we suggest an algorithm that permits quick computation of contrasting patterns by exploiting clever traversal and pruning methods. We showcase the potential of contrasting patterns on a wide range of artificial and real-world datasets.
