Improving local search algorithms for clique relaxation problems via group driven initialization

Rui Sun, Yiyuan Wang, Minghao Yin

Abstract: Clique relaxation problems are important extension versions of the maximum clique problem with extensive real-world applications. Although lots of studies focus on designing local search algorithms for solving these problems, almost all state-of-the-art local search algorithms adopt a common general initialization method. This paper develops a general group driven initialization method for clique relaxation problems. The proposed method uses two kinds of ways to divide vertices into some subgroups by using the useful information of the search procedure and the structure information of a given instance and then constructs a good initial solution by considering the generated group information. We apply the proposed initialization method to two clique relaxation problems. Experimental results demonstrate that the proposed initialization method clearly improves the performance of state-of-the-art algorithms for the clique relaxation problems.