Aim/PurposeThis study addresses the challenge of understanding and predicting tourist decision-making in the AI era by integrating sentiment, credibility, and contextual signals from social media into a unified and actionable framework. It seeks to move beyond raw user-generated content toward trustworthy, decision-ready insights that can guide destinations, platforms, and travelers.

BackgroundTourism analytics often treat online sentiment at a surface level, overlooking emotional nuance, content credibility, and real-world context. To bridge this gap, the Affective Understanding, Reliability, and Outcome-driven Recommendation Architecture (AURORA) offers a sentiment-driven, context-aware system that captures emotions, filters unreliable information, and models how travelers make choices across different decision stages.

MethodologyAURORA processes multimodal public data such as reviews, social posts, and event feeds. It combines aspect-based sentiment and emotion analysis with credibility assessment and contextual modeling. A Bayesian state-space model tracks traveler decision stages, while uplift modeling identifies when and where interventions are most effective. The analysis tested AURORA on a large-scale Booking.com dataset spanning 18–24 months, which captured seasonal variation.

ContributionAURORA introduces a next-generation, sentiment-based decision framework that unites emotional understanding, trust evaluation, and context sensitivity in tourism analytics. The paper demonstrates how combining textual and behavioral data yields measurable insights that are both theoretically grounded and practically deployable for decision support.

FindingsAspect‑level sentiment on core attributes – particularly safety, cleanliness, and value – emerges as the strongest predictor of traveler attitudes and booking propensity, with emotional intensity amplifying these effects and social proof showing diminishing marginal returns at high volumes. Credibility and provenance filters materially reduce noise from low‑quality or manipulated content, and hybrid models that combine textual embeddings with structured metadata outperform polarity‑only baselines for predicting high‑intent behaviors. Practitioners should therefore prioritize real‑time monitoring of high‑elasticity aspects, elevate aspect‑rich, high‑credibility content in ranking and messaging, and time interventions to the traveler’s decision stage to maximize incremental lift rather than mere engagement.

Recommendations for PractitionersPrioritize real-time monitoring of high-elasticity aspects (e.g., safety, crowding) to inform campaigns.

Elevate high-quality, aspect-rich user content in ranking algorithms.

Time interventions to decision-journey stages when cues are most impactful.

Recommendation for ResearchersExtend the framework to non-English, multilingual contexts with localized aspect taxonomies.

Investigate causal pathways between exposure to specific sentiment cues and actual booking behavior.

Explore the integration of multimedia sentiment (image/video emotion) in tourism decision models.

Impact on SocietyEnables more transparent and trustworthy tourism information ecosystems, empowering travelers to make better decisions and helping destinations manage perception during crises. Fosters healthier digital tourism discourse by amplifying authentic, credible voices.

Future ResearchBuilding on AURORA, the next phase should validate and extend the architecture across languages, cultures, and media. This includes developing localized aspect taxonomies and multilingual encoders, running causal field experiments that link exposure to specific sentiment cues with booking behavior, and integrating multimedia emotion signals (images and video) alongside privacy‑preserving, on‑device analytics. Together, these steps will make AURORA more robust to cross‑cultural expression, resilient under distribution shifts, and practical for real‑time, privacy‑aware deployment in diverse tourism ecosystems.