Evolving to recognize high-dimensional relationships in data: GA operators and representation designed expressly for community detection.

We present a new algorithm for network clustering based upon genetic algorithm methods to optimize modularity. The algorithm proposes an innovative, more abstract representation, along with newly designed domain-specific genetic operators. We then analyze the performance of the algorithm using popular real-world data sets taken from multiple domains. The analysis demonstrates that our algorithm consistently finds high quality or even optimal solutions without any a priori knowledge of the network or the desired number of clusters. Furthermore, we compare our results with five previously published methods and yield the highest quality for the largest of the benchmark datasets.