Clustering of Correlated Documents into Designated Number of Clusters: A Practical Approach

We consider a complete graph to cluster the vertices into k-clusters, where each edge of the graph is labeled either as “+” or “–”. “+” denotes that vertices incident to the edge are mutually related and “–” edge denotes that the vertices incident to the edge are mutually unrelated. The goal of the clustering is to place vertices into k clusters, where documents are clustered with maximally related items. That is, clustering should maximize the agreements (“+” edges inside the clusters and “–” edges between the clusters), or equivalently minimizes the disagreements (“–” edges inside the clusters and “+” edges between the clusters). We give a simple algorithm for the maximizing the agreements and test the success of the algorithm. We compare approach with Bansal et. al’s approach proposed in [1]., and conclude that complicated algorithms that have exponential run time are not practical.