Community detection plays a key role in uncovering group structures in networks, but its misuse can lead to privacy risks by exposing sensitive relationships. As a proactive defense, community hiding seeks to perturb the network structure to reduce the effectiveness of community detection algorithms. Given the NP-hard nature of this task, genetic algorithms (GAs) widely used due to their robust global search capabilities. However, existing methods lack effective guidance when dealing with vast solution spaces, resulting in inefficient exploration and suboptimal obfuscation outcomes. To address this, we propose Network-Topology-Combined Community Information Hiding Algorithm (TCHA), a novel GA-based method that leverages node similarity information via node embedding to guide perturbations more effectively. TCHA introduces a multidimensional mutation operator that combines coarse-grained and fine-grained mutation strategies. These fine-grained mutations are performed in the embedding space and decoded back to graph edits, enabling more precise and topologically-aware perturbations. To evaluate the efficiency of this process, we introduce a novel metric based on expected path length within a mutation transition graph, offering deeper insight into evolutionary search dynamics. Experiments on six real-world networks demonstrate that TCHA achieves an average modularity reduction of 31.92% and an average normalized mutual information of 0.6447, outperforming baselines such as Q-Attack, NEURAL, and DICE. These results confirm the superiority of the embedding-guided fine-grained mutation strategy in enhancing community hiding effectiveness.