Compared with group decision making (GDM), it is more difficult to reach a consensus in large-scale group decision making (LSGDM) owing to the large number of decision makers (DMs). Moreover, studies on LSGDM under social trust networks are obviously fewer than those on GDM, and rarely utilize opinion dynamics to analyze the interaction of DMs. Therefore, in the context of multi-criteria large-scale group decision making (MCLSGDM), an MCLSGDM consensus decision framework and a bounded confidence-based consensus optimization model are proposed. First, a trust propagation method considering the relative importance of the trust degrees (TDs) is proposed. Then, a two-stage process of obtaining DMs’ opinions based on opinion dynamics is developed to analyze the interaction of DMs before clustering. Furthermore, a new method for determining the heterogeneous weights of DMs and subgroups is discussed. Finally, to consider the adjustment willingness of DMs, this study proposes a two-stage optimization consensus model based on bounded confidence. In addition, a numerical example is used to further elaborate the above methods and models, and highlight their rationality and superiority through a series of simulation experiments and comparative analysis.