Information cascades are recognized as a major factor in catastrophic social network phenomena. Network structure and user behavior have important impacts on information cascades. In particular, homogeneity often becomes a barrier to the spread of new things, making it difficult to adopt new ideas and behaviors from outside closely connected com-munities. In this paper, we use cluster structure to precise densely connected community, and there exist the similar tendencies of behaviors between individuals and their neighbors in the cluster. A model of information cascades is proposed to emphasize individual deci-sion-making based on the choices of their neighbors and the content of spreading. On the basis of this model, an efficient method is developed for restraining information cascades by reconfiguring clusters with links logically removed (termed CLLR). By limited links with high betweenness being logically removed, the CLLR method enhances the cluster density of a network, thereby effectively block information cascades. Experimental results show that the CLLR method significantly reconfigure the cluster structure with limited links removed, efficiently restrain the speed and scope of information cascades in mobile social networks.