<p>Most of the existing publications shedding light on large-scale group consensus-reaching focus on the design of opinion adjustment rules based on an exogenous decision-maker social network. However, this perspective overlooks the fact that the structure of social interactions itself can serve as an active decision variable, which may limit the achievable level of group consensus. To address this issue, this paper proposes a novel framework for ordinal-cardinal consensus-reaching through active fine-tuning of social network structures. Instead of directly modifying individual opinions, the proposed approach dynamically adjusts interaction links among decision-makers, thereby influencing opinion evolution through structural interventions. Building on the psychological theories of locus of control and compensatory control, we model how changes in interaction structure affect decision-makers’ self-confidence and trust relationships, which in turn shape their willingness to participate in the consensus process. In this way, our study captures the collaborative dynamics of large-scale group decision-making by explicitly representing how human decision-makers respond to artificial intelligence (AI)-assisted coordination of their interaction patterns. We further formulate this network structure fine-tuning problem as a Markov decision process and develop a multi-objective reinforcement learning method based on multi-gradient ascent and trust region policy optimization. This provides an AI-enabled decision-support approach for facilitating groups in reaching a consensus. Extensive simulations validate the effectiveness and robustness of the proposed framework, highlighting the potential of structure-aware and behavior-aware intervention mechanisms for large-scale group decision-making.</p>