As a theoretical method for solving complex real-life problems, group decision-making (GDM), along with the rapid development of artificial intelligence technology, has led to intricate and complex decision-making situations. This has contributed to the rise and rapid evolution of complex large-scale GDM (LSGDM). In the LSGDM process, the reasonable grouping of decision-makers (DMs) and reaching a consensus are the core links to obtain the optimal decision-making scheme, and these rely heavily on mutual trust among DMs. However, in reality, not all trust is real and effective, and pseudo-trust is a common phenomenon. As such, identifying and managing pseudo-trust behavior by DMs has become a challenge. This study investigates the influence of pseudo-trust on DMs' dimensionality reduction and consensus process and proposes a social network LSGDM method based on feature selection and pseudo-trust behavior. Specifically, it proposes a leader feature selection based on a dual trust relationship to address the efficiency and rationality challenges in large-scale DMs' dimensionality reduction. Through this study, we provide a clear concept of pseudo-trust behavior and create a quantitative assessment system. Furthermore, an adaptive consensus model based on pseudo-trust behavior is constructed to achieve its effective identification and management. Finally, the effectiveness, practicability, and superiority of the proposed method are proven by selecting real-world cases from the UCI database, combined with experimental and comparative analyses.

Computational intelligence is increasingly applied to complex decision-making challenges, leveraging its data analysis prowess. Hybrid human-artificial intelligence models enhance the grasp of intricate social behaviors, offering valuable insights for social computing and behavior modeling. Within this landscape, large-scale group decision-making (LSGDM) emerges as an invaluable asset for navigating intricate decision-making scenarios. LSGDM enlists the expertise of individuals who articulate their preferences via fuzzy preference relations (FPRs) that abide by additively consistent principles. Its ascendancy is underscored by its applicability and relevance in confronting multifarious decision-making conundrums. In the realm of LSGDM, machine learning methodologies, such as cluster analysis, are deployed to streamline decision-making procedures, particularly when confronted with inherent complexities. The consensus reaching process (CRP) serves as the cornerstone, ensuring that decision-makers (DMs) converge on a unified verdict. Consequently, comprehensive exploration of cluster analysis and CRP assumes a pivotal role in elevating the effectiveness of LSGDM. To further augment LSGDM, this study leverages a three-way clustering (TWC) approach grounded in adaptive fuzzy$c$-mean clustering. This stratagem categorizes DMs into discrete subgroups. Moreover, a consensus metric, embracing both cardinal and ordinal consensus considerations, is established. This metric serves as the foundation for computing DMs' intra-group weights and group weights. Moreover, this paper introduces a feedback mechanism imbued with identification and modification rules (directional rules). It incorporates a modification function that takes into account the consensus threshold, DMs' regret psychology and the consensus level. This modification function methodically derives modification parameters for the spectrum of DMs. Lastly, the viability and effectiveness of the LSGDM methodology proffered in this paper are substantiated via meticulous simulation and comparative analyses.