In the realm of social networks, community structures play a pivotal role in influencing the dissemination of opinions, with highly influential nodes within these communities serving as the linchpin for shaping collective opinion. However, previous research on opinion propagation and control related to community detection often overlooked the impact of the initial opinion distribution and seldom delved into strategies for steering the opinions of these influential nodes. Against this backdrop, this paper introduces an adaptive opinion control method based on two-stage community detection. This method comprises three key components: 1) Improved Opinion Dynamics Model: The difference between the influence of nodes on the opinions of other nodes and the firmness of their own opinions is considered. The nodes in the network are divided into authoritative nodes and ordinary nodes, and an improved opinion dynamics model is established. 2) Two-Stage Community Detection Based on Similarity: In the first stage, considering the structural similarity and opinion similarity in the network, the social network with initial opinions is pre-divided. The Louvain algorithm is then used in the second stage based on the results of the first stage of partition. Several communities with close internal opinions and tight structures are finally partitioned. 3) Adaptive Opinion Control Targeting Authoritative nodes: Considering the community-specific and dynamic nature of opinions, this method can realize the sub-community and gradual guidance of public opinion. Experimental results on four distinct social network datasets affirm the effectiveness of each model, aligning well with the real world.