Summary: Large-scale group decision making is common in real-world scenarios, yet it involves two critical issues: (1) clustering individuals into subgroups according to specific criterion, and (2) facilitating any subsequent consensus reaching process. This paper presents a novel approach to address these challenges. Firstly, a set of transformation rules is proposed to convert heterogeneous judgments expressed by individuals into a homogeneous preference form. These judgments can be classified from two aspects: direct or indirect assessments, and fuzzy set or linguistic term set schemes. Subsequently, a group clustering method is introduced to classify individuals into subgroups, considering both of their preferences and social relations. The clustering method incorporates the measures of opinion divergence among individuals within the group and social network analysis techniques comprehensively. Finally, an adaptive two-stage group consensus measurement and adjustment method is proposed. The first stage employs a centralized mechanism within each subgroup, aiming to achieve intra-subgroup consensus. The second stage employs a democratic mechanism among different subgroups, focusing on inter-subgroup consensus. The effectiveness and rationality of the proposed method are demonstrated through an illustrative example and comparative analysis with state-of-the-art methods. The findings highlight the usefulness of the proposed method in addressing real-world decision-making problems within large-scale group contexts.