The inverse graph model for conflict resolution (GMCR) aims to help stakeholders achieve the optimal preferences that can make the desired state an equilibrium. Common interests often drive stakeholders to form interest coalitions. Existing studies on inverse GMCR are primarily conducted in numerical environments and tend to overlook the consensus issues within interest coalitions. In reality, however, preference divergences arising from stakeholders’ diverse knowledge backgrounds often hinder interest coalitions from achieving greater common benefits, and their decision-making processes typically take place in linguistic environments characterized by personalized individual semantics (PISs). To address this, this study proposes a consensus-based inverse GMCR that incorporates PISs, and employs linguistic distribution assessments (LDAs) to represent stakeholders’ preferences. Specifically, this study first introduces a consensus-driven PIS model with historical data, which aims to address stakeholders’ PISs by utilizing historical conflict data of interest coalitions. Subsequently, a minimum-adjustment-based inverse preference optimization model with bounded confidence is proposed to assist stakeholders in attaining optimal preferences that render the desired state an equilibrium. By incorporating minimum adjustment and bounded confidence, the proposed model reduces adjustment costs and enhances stakeholders’ willingness to revise their preferences. At the same time, the model ensures both group consensus among stakeholders and individual consistency. Finally, the feasibility and effectiveness of the proposed method are demonstrated through a hypothetical case, sensitivity analysis, comparative analysis, and a discussion.