Robust Bi-Level Multi-Scale Group Consensus: Integrating Spatiotemporal Online Review Influence Analysis for Portable Consumer Electronics in Smart City Tourism

In smart city tourism (SCT), tourists share scenic site preferences via portable consumer electronics (PCE), creating diverse opinions. This complicates destination selection for future tourists, necessitating group decision-making (GDM) to aggregate these perspectives into useful recommendations. However, existing GDM methods face challenges such as inadequate comment quality screening, insufficient spatiotemporal feature modeling, personalization-ignorant clustering, and unstable consensus-reaching process (CRP). To address these gaps, a robust bi-level consensus-reaching method is proposed, which clusters decision-makers (DMs) based on personalized traits extracted from online reviews on social networks (SNs). First, online reviews generated via PCE are collected from target platforms, with influential reviews selected via human-machine interaction. Second, sentiment analysis via eXtreme language model (XLNet) generates time-series evaluation matrices. Subsequently, multi-scale information systems (MSISs) are used for spatial modeling, the mapping data to a higher-dimensional structural space. Third, the stability and consistency measure DMs’ abilities, while responsibility and agreeableness capture their attitudes. Fourth, these traits are integrated into CRP to construct a multi-granularity bi-level consensus model. For the subgroup disagreement, a hybrid optimization and rule-based approach based on Stackelberg game ensures robust consensus by minimizing cost and maximizing fairness (MCAMF). Finally, a case study on scenic site recommendations validates the method’s feasibility.