Correlation Clustering with Noisy Partial Information

Abstract

In this paper, we propose and study a semi-random model for the Correlation Clustering problem on arbitrary graphs G. We give two approximation algorithms for Correlation Clustering instances from this model. The first algorithm finds a solution of value (1+ ) optcost + O_(n^3 n) with high probability, where optcost is the value of the optimal solution (for every > 0). The second algorithm finds the ground truth clustering with an arbitrarily small classification error (under some additional assumptions on the instance).

Tasks

Reproductions