In the last years, the study of complex networks grows rapidly and search of tightly connected groups of nodes, or community detection, has proved to be a powerful tool for analyzing the real systems. Randomized algorithms are effective for detecting communities but there is no set of optimal parameters that makes these algorithms create a good partitions into communities for every input complex network. In this paper we consider two randomized algorithms and, based on the stochastic approximation, propose two new adaptive modifications that adjust parameters to the input data and create a good partitions for wider range of input networks.