A Heterogeneous Hybrid Re-Sampling (HHR-S) Enhanced Emoji-Aware Generative Expert System for Aspect-Level Tourism Analytics

Online tourism reviews are a vital source of experiential intelligence, yet conventional sentiment analysis pipelines face challenges such as severe class imbalance, limited use of affective signals, and unconstrained generative outputs that reduce interpretability and reliability. This study introduces a heterogeneous hybrid re-sampling and emoji-aware generative expert system for aspect-level tourism analytics. The framework incorporates three key mechanisms: (1) structured emoji polarity encoding as an affective knowledge representation layer; (2) imbalance-aware minority synthesis with calibrated decision-boundary optimization to improve classification under skewed data; and (3) taxonomy-guided generative inference for controlled aspect-level intelligence extraction. Unlike previous methods that treat emojis as auxiliary tokens or use large language models without structural constraints, the proposed architecture formally integrates affective and semantic information to enhance robustness and explainability. An experimental evaluation of 13,501 real-world tourism reviews shows that the imbalance-aware configuration achieves 95.94% accuracy and statistically significant improvements in Macro-F1 over text-only and non-resampled baselines. The generative module also demonstrates superior macro-level discrimination over transformer-based aspect-based sentiment analysis baselines. By combining affective computing, imbalance-robust learning, and taxonomy-constrained generative reasoning within a deployable decision-support system, this work advances tourism analytics toward interpretable, stable, and operationally reliable intelligence solutions.