Dynamic Self-Confidence Management for Multi-Granularity Probabilistic Linguistic Group Decision Making with Fuzzy Social Networks and Overconfidence Correction
Main article
Abstract
Group decision-making (GDM) in contemporary industrial settings requires simultaneously addressing the heterogeneity of decision-maker (DM) preference representations, the dynamic nature of DM confidence levels, and the trust-mediated social influences that shape consensus formation. Probabilistic linguistic term sets (PLTSs) offer a flexible mechanism for capturing nuanced preference information with associated probability distributions, yet existing GDM frameworks do not adequately address multi-granularity PLTS environments where different DMs employ different linguistic scales. Furthermore, the pervasive phenomenon of DM overconfidence---the tendency to assign excessively high self-confidence levels that distort aggregated group opinions---remains largely unaddressed in probabilistic linguistic GDM. This paper proposes a dynamic self-confidence management framework for multi-granularity probabilistic linguistic GDM that integrates three novel components: a multi-granularity PLTS (MG-PLTS) transformation method for unifying heterogeneous linguistic scales through a common semantics-preserving mapping; a fuzzy social network (FSN) for modeling trust-based influence relationships among DMs; and an overconfidence detection and correction mechanism that identifies and adjusts inflated confidence levels based on historical decision accuracy and peer trust assessments. The dynamic confidence update rule adjusts individual DM weights in proportion to their relative consensus contribution and FSN-mediated trust scores, accelerating convergence. Empirical validation using online review sentiment data for industrial supplier evaluation demonstrates that the proposed framework achieves consensus (threshold 0.90) in 4.2 rounds on average for a 5-DM group, compared to 7.8 rounds for basic GDM, 5.9 rounds without FSN integration, and 5.1 rounds without overconfidence correction. Application to a real-world green supplier selection problem confirms superior decision quality and DM satisfaction.
