Comparative Study on Optimization Methods for Correlation Clustering

Abstract

Correlation clustering is an optimization problem that aims to create partition of data based on pairwise similarity coefficients that represents the level of similarities between data observations. The thesis focused on the maximization of agreements version of the problem in which to find clustering of data where the data that belong to the same cluster have maximized agreements. The thesis aims to give more details on how different methods are used for correlation clustering problem, how they perform and what are the similarities between these methods. The well known linear programming methods as well as simple iterative algorithms are compared in terms of their runtime and correctness.