<p>Consensus formation is pivotal in large-scale group decision-making, significantly influenced by the structure of social networks, thereby drawing substantial attention to large-scale consensus-reaching in social networks (LSC-SN). However, existing clustering methodologies for organizing decision-making experts often encounter challenges when applied to extensive networks and frequently overlook the inherent social network structures. To address these limitations, this study first introduces an evaluation similarity factor to enhance modularity. Then, we propose an improved Louvain algorithm to increase the efficiency and adaptability of network segmentation. In the consensus-reaching process (CRP), a novel Collaborative Similarity Index (CSI) is developed to regulate the cooperative behavior of decision-makers and prevent over-adaptation. Building upon this, a novel approach is introduced to simultaneously determine both decision-maker weights and community weights, while also considering the degrees of consensus within and between communities. Additionally, a dual-domain reference consensus feedback model based on CSI is constructed, integrating individual opinions, inter-domain references, and inter-domain opinion references to promote fairness. Finally, the validity and feasibility of the proposed methodology are substantiated through empirical research and comprehensive analysis.</p>