Collaborative Intelligence for Group Decision Problems Under Conflicting Preferences and Uncertainty

Group decision making problems in complex socio-technical areas are characterized by conflicting preferences among stakeholders and significant uncertainty. These dimensions are usually tackled in isolation by the existing decision support approaches. This study proposes a collaborative intelligence framework that combines probabilistic linguistic term set encoding, conflict-derived influence weighting, TOPSIS based on expected probabilistic-linguistic indices, and preference revision. The reported experiment executes the deterministic pipeline and does not exercise the recommendation component. The probabilistic encoding rule converts each linguistic judgment into a probability-weighted term set. The framework measures pairwise divergence with a normalized Manhattan distance. The influence allocation adjusts expert contribution under equal prior credibility. The TOPSIS procedure based on expected probabilistic-linguistic indices ranks the weighted assessments. The framework is assessed on the APOLLO dataset of ten experts, twenty-five risks, and four criteria. The proposed framework reaches a consensus of 0.9455 after 19 feedback iterations, compared with 0.8793 for the equal-weight common-metric baseline on the same metric. The configured stopping threshold of 0.95 is not reached within the allowed iteration horizon. The run terminates at the iteration cap. The ranking agreement with the equal-weight common-metric baseline equals 0.9892 under the Spearman correlation. The preference revision loop executes 19 feedback cycles because the initial consensus falls below the configured threshold. No LLM call occurs in the reported run. The consensus improvement thus denotes the deterministic pipeline and not any of the recommendation elements. This attribution is supported by the ablation study. The results confirm the reproducibility of the deterministic pipeline.