Abstract
The rapid expansion of digital tourism and traveler data presents a significant opportunity for personalization, yet existing systems struggle to adapt to ever-changing tourist preferences within context. This study proposes a novel multi-dimensional artificial intelligence (AI) framework leveraging advanced machine learning for hyper-personalized, real-time tourism experiences. The architecture integrates transformer-based neural networks, reinforcement learning agents, and ensemble methods to process heterogeneous data streams, including user behavior, social media, environment, time, and culture. The method used was a hybrid method combining the collaborative filtering recommendation system along with the content-based recommendation system along with improvement using real-time adaptation and optimization algorithm. As more and more tourists started utilising the tech support service, the framework was validated through a large deployment at various tourism ecosystems. Also, the analysis included the recognition of 2.3 million users interactions and 450000 during service transactions in 15 nations. Results show a significant increase in recommendation accuracy for single-item recommendations (78% improvement), personalization accuracy (91.2%), and satisfaction score (65% increase) compared to conventional systems. The system based on AI take less than 0.4 s for a response while reducing computing overhead by 42% through algorithm. This study introduces an innovative approach to intelligent tourism systems that utilizes real-time contextualized information and predictive behavior models to improve tourist satisfaction. Innovative mathematical frameworks for modelling tourism preferences and new cross-cultural adaptation techniques. The practical implications are better experience for customers, revenue maximization for tourism operators and sustainable tourism development through optimal allocation of resource.
Plain Language Summary
This research explores how artificial intelligence (AI) can create highly personalized travel experiences. Just as streaming services suggest movies you might like, we developed a smart system that learns your travel preferences to recommend destinations, activities, and services that are a perfect fit for you. Unlike standard travel websites, our system doesn't just look at your past bookings. It intelligently considers real-time factors like local weather, current events, and even cultural trends. It also understands that your preferences can change during a trip. By analyzing a wide range of information, including your online behavior and feedback, the system continuously adapts its suggestions. We tested this technology with millions of users across 15 different countries. The results were significant. The system was over 91% accurate in making relevant recommendations, far outperforming current travel platforms. Users were much more satisfied, with satisfaction scores rising by 65%. For businesses like hotels and tour operators, this personalization led to a 38% increase in revenue, as customers found and booked options they truly wanted. A key achievement is the system's cultural intelligence. It successfully tailors recommendations for travelers from different parts of the world, ensuring suggestions are not only interesting but also culturally appropriate. This leads to more respectful and enjoyable tourism. In short, this study shows how smart, adaptive AI can transform travel planning. It benefits everyone: travelers get a more tailored and satisfying experience, while tourism businesses operate more efficiently and profitably. This technology paves the way for a smarter, more responsive future for the global tourism industry.
Keywords
Introduction
The tourism sector has undergone a unique digital transformation in recent years. This has altered the discovery, plan, and experience of the travels. The databases generated by global tourism are growing exponentially. With a staggering 4.8 billion daily digital interactions on tourism platforms such as OTAs and hotels, there is an urgent need for sophisticated personalization mechanisms to support industry sustainability and competitive advantage. More and more contemporary travelers wish to have personalized experiences, rather than a common standard one. This opens the door to a large-scale, AI-powered solution to process complex, multidimensional traveler preference patterns in real time (Ghesh et al., 2024).
Recent developments in artificial intelligence have prompted platforms in the tourism technology sector to innovate. The personalization system based on fuzzy logic has been successfully applied in research to enhance user experience through intelligent recommendations (Chrysafiadi et al., 2025). ATLANTIS (2023) introduced innovative tourism services that incorporated deep learning techniques to build neural network architectures for personalization. This work has established an initial reference point. In the same vein, Kontogianni et al. (2024) developed comprehensive innovative tourism frameworks, while Aliyah et al. (2023) explored the integration of AI and IoT for destination management. Fang-ming and Haotian (2022) developed a machine learning-based cloud IoT platform to provide intelligent information services. Li and Zhang (2022) designed innovative tourism platforms from an AI perspective. Those establish the technology foundation for innovative personalization systems. Recommendation systems have been studied across numerous paradigms and in evolution. Bobadilla et al. (2013) provided a detailed survey of the recommendations system methodologies and established benchmark methodologies for collaborative filtering and content-based filtering. Ricci et al. (2015) explained fundamental design issues in recommendation systems, while Braunhofer and Ricci (2017) developed techniques for selectively acquiring contextual information for travel recommendations. García et al. (2018) developed microservice-based architectures for tourism platforms, while Alvarado-Uribe et al. (2017) implemented intelligent point-of-interest recommendation algorithms. Nilashi et al. (2022) addressed data sparsity issues using hybrid fuzzy approaches. Together, these works are advancing the sophistication of tourism recommendation systems. Intelligent tourism systems must take context into account. Feng et al. (2014) were the first to investigate the context-aware service mechanism, which served as a theoretical basis for characterizing environmental and temporal features. Khallouki et al. (2018) developed an ontology-based context-awareness framework for a recommendation system, while Jorro-Aragoneses et al. (2017) implemented context-aware leisure plan recommendations. According to Smirnov et al. (2014), the innovative space-based tourist recommendation system proves the importance of personalization frameworks. Grasping the intricacies of personalization is an important research avenue. Neidhardt and Werthner (2019) explored the challenges involved in personalizing the tourism experience, specifically how temporal preferences change over time. Ghesh et al. (2024) conducted systematic literature reviews on AI adoption in customer experiences, and Ku and Chen (2024) examined the link between AI innovation and service usage intention. Semwal et al. (2024) investigated AI-powered personalization, along with the integration of emotional intelligence, to develop a multidimensional approach to enhance the consumer experience. An in-depth analysis of the industry reveals AI implementations ranging from general to specific. Filieri et al. (2021) identified the success factors and implementation challenges of successful European AI tourism start-ups. Kong et al. (2023) highlight the studies in the field of AI by examining the last thirty years of research from AI and hospitality and tourism. Pillai and Sivathanu (2020) examined how people use chatbots, and Grundner and Neuhofer (2021) highlighted the benefits and drawbacks of using AI in a destination experience. Tourism can benefit from advanced analytical abilities. According to Kumar and Zymbler (2019), a machine learning approach can be utilized for customer satisfaction analysis. Ye and Huang (2022) also implemented deep learning-based neural networks for analysing psychological behavior. Studies by Zulaikha et al. (2020) contributed to customer predictive analytics methodologies; He et al. (2024) exploited large language models for demand forecasting; and Mo et al. (2024) explored the spell-out of AI awareness and its impact on service performance adaptivity.
Recent advances in artificial intelligence have introduced large language models (LLMs) and generative AI as emerging tools for tourism personalization. LLM-based systems have been applied to conversational travel assistants, automated itinerary generation, and natural language interaction, enabling users to express preferences more flexibly and receive human-like responses. Generative AI techniques have also been explored for producing personalized travel narratives, recommendations, and destination descriptions, enhancing user engagement and interaction quality. These approaches represent a significant step toward more intuitive and interactive tourism platforms.
However, existing generative AI–driven tourism systems also exhibit notable limitations. Most LLM-based applications rely on static or periodically updated knowledge bases and lack robust mechanisms for real-time preference evolution, fine-grained contextual adaptation, and multi-stakeholder optimization. In addition, their high computational cost and dependency on large-scale centralized models pose challenges for real-time deployment, scalability, and data privacy in cloud environments. Furthermore, current generative approaches often prioritize linguistic fluency over optimization accuracy, offering limited control over recommendation objectives such as revenue optimization, sustainability, and cross-cultural adaptation. These limitations highlight a critical research gap, motivating the need for a multidimensional AI framework that integrates deep learning, reinforcement learning, and optimization mechanisms to deliver real-time, context-aware, and stakeholder-balanced tourism personalization beyond the capabilities of existing generative AI solutions.
Existing tourism personalization systems are limited in their technical capabilities, as the literature reveals. Currently, recommendation algorithms primarily rely on static user profiles and thus cannot reflect the dynamic, contextual nature of traveller preferences, which change throughout the various stages of the journey (Braunhofer & Ricci, 2017; Neidhardt & Werthner, 2019). A second drawback is that existing solutions fail to connect heterogeneous data sources, including behavioral patterns, social media sentiment, environmental factors, and cultural preferences, to personalize on a large scale (Kontogianni et al., 2022). Third, the capability of real-time adaptation is still underdeveloped. Most systems have significant processing times, which significantly impact the quality of the user experience (García et al., 2018). Fourthly, the adaptability mechanisms used to address the different international traveler populations are not sophisticated enough. (Khallouki et al, 2018) In the end, current frameworks do not offer adequate multi-stakeholder optimization frameworks that jointly optimize traveller satisfaction, operator profitability, and sustainable tourism development (Ghesh et al., 2024).
Recent studies consistently demonstrate that real-time cybersecurity systems benefit substantially from hybrid AI/ML architectures that combine anomaly detection, behavioral analytics, and adaptive learning. Real-time intrusion detection frameworks employing deep autoencoders with lightweight classifiers such as K-Nearest Neighbors have proven effective for identifying both known and unknown attacks under strict latency constraints, particularly when integrated with live packet capture tools and modular alerting mechanisms (Vishrutha & Nagaraja, 2024). Complementing this, comprehensive AI-driven threat detection surveys show that machine learning techniques, ranging from anomaly detection and traffic analysis to IDS/IPS and user behavior analytics, outperform traditional rule-based approaches in accuracy and responsiveness, especially in large-scale, heterogeneous data environments (Chen et al., 2024). The integration of Big Data analytics further enhances these systems by enabling continuous learning from diverse data sources and early prediction of emerging threats, while explainable AI components improve transparency and operational trust (Balusamy et al., 2024).
Beyond external threats, recent literature emphasizes insider threat detection and adaptive trust management as critical components of secure, cloud-based systems handling sensitive behavioral and sentiment data. AI-based insider threat frameworks leverage supervised and unsupervised models, including Random Forests, SVMs, Isolation Forests, RNNs, and Autoencoders, to establish behavioral baselines and detect subtle deviations indicative of malicious intent, while NLP techniques extend detection to unstructured data such as emails and chat logs (Parthasarathy et al., 2024). In parallel, AI-driven zero-trust architectures apply continuous user behavioral analytics to dynamically reassess access privileges based on contextual signals such as device attributes, location, and access patterns, reinforcing security in real-time cloud environments (Ahammed & Labu, 2024). From an adoption and governance perspective, empirical studies grounded in the UTAUT model highlight that trust, regulatory compliance, and perceived risk strongly influence organizational uptake of ML-based cybersecurity solutions (Olaniyi, 2024). Finally, human-centric and bias-aware perspectives from interactive and generative information retrieval research underscore the importance of responsible AI design, interpretability, and cognitive bias mitigation when deploying intelligent systems that process sensitive user data at scale (Liu & Azzopardi, 2025).
Recent interdisciplinary research provides strong evidence that AI- and ML-driven detection and optimization techniques developed in cloud, IoT, financial, and industrial domains are directly applicable to security-sensitive tourism systems. In cybersecurity and industrial monitoring, hybrid deep learning models and optimization-driven architectures are widely used for real-time anomaly detection, adaptive control, and performance optimization under dynamic conditions. These systems demonstrate how advanced learning algorithms can simultaneously achieve accuracy, robustness, and low latency, properties that are equally critical for tourism platforms processing continuous streams of behavioral and transactional data.
In parallel, extensive work on data security in cloud and IoT environments highlights the necessity of integrating privacy-aware analytics, secure model deployment, and real-time monitoring when handling sensitive user data. Finance and industrial AI systems, which operate under strict regulatory and operational constraints, further illustrate how machine learning can be deployed securely for high-frequency transactions, fraud detection, and decision optimization. These domains share key characteristics with intelligent tourism platforms, including real-time processing requirements, heterogeneous data sources, and the need to balance performance with trust, privacy, and compliance. By grounding the proposed framework in these established research streams, the manuscript aligns tourism personalization with proven AI methodologies for secure, real-time, and scalable system deployment.
While prior research in smart tourism personalization has explored collaborative filtering, content-based recommendation, and context-aware services independently or in limited hybrid forms, this study advances the field by introducing a unified, multidimensional AI framework that explicitly models preference evolution, real-time adaptation, and multi-stakeholder optimization within a single architecture. Unlike existing systems that rely on static user profiles or isolated contextual features, the proposed framework integrates transformer-based sequential modeling, reinforcement learning–driven decision optimization, and ensemble learning to dynamically capture how tourist preferences change over time and across contexts.
In contrast to recent studies that focus primarily on conversational agents, generative recommendation interfaces, or single-objective optimization, this work contributes a conceptually distinct approach by formalizing tourism personalization as a multi-objective optimization problem that jointly considers user satisfaction, economic performance, and sustainability. Methodologically, the framework introduces mathematically grounded preference evolution models and Pareto-optimal decision strategies, enabling adaptive personalization under real-time, cross-cultural, and large-scale deployment conditions. These contributions position the study beyond an implementation example, offering a transferable theoretical and methodological foundation for next-generation intelligent tourism systems.
This study is grounded in established theories from tourism and information systems research to provide a coherent theoretical foundation for AI-driven personalization. First, Experience Economy Theory posits that value in tourism emerges from personalized, memorable experiences rather than standardized services, directly motivating adaptive recommendation mechanisms. Second, Context-Aware Information Systems theory explains how systems can sense, interpret, and respond to situational factors, such as time, location, and social context—to improve decision relevance. Third, Expectation–Confirmation Theory provides a basis for understanding customer satisfaction and continued usage by linking perceived performance with prior expectations. Together, these frameworks justify the integration of dynamic preference modeling, contextual intelligence, and adaptive optimization in the proposed system.
Guided by these theoretical foundations, this study addresses the following research questions:
This paper seeks to address identified limitations by making four key theoretical and practical contributions. To enable high-accuracy, efficient processing of real-time heterogeneous data streams and present a unique multidimensional AI framework that leverages enhanced transformer-based deep neural networks, RL agents, and an ensemble method and develop cutting-edge mathematical models for the evolution of dynamic preferences. This takes into account various temporal, contextual, and cross-cultural factors. Thus, this is achieved using innovations in machine learning. The third one is to utilize innovative multi-stakeholder optimization approaches to improve customer satisfaction, operational efficiency, and sustainability. In the fourth step, the effectiveness of different personalization systems is validated in international tourism markets. Substantial improvements in recommendation accuracy, response time, and customer satisfaction are demonstrated compared to existing systems.
Methodology
The framework also integrates machine learning systems, contextual intelligence systems, and real-time optimization systems to attain hyper-personalized tourism experiences. The approach consists of four major stages: system architecture design, algorithm development, mathematical modeling and validation. The framework considers the findings and suggestions of the up-to-date recommendation frameworks and presents innovative strategies of dynamic preference modeling and multi-stakeholder optimization (Bobadilla et al., 2013; Ricci et al., 2015).
In this study, contextual intelligence refers to the system’s ability to interpret and utilize situational factors, such as time, location, environmental conditions, and social context, to dynamically adjust recommendations beyond static user profiles. Cross-cultural adaptation techniques denote algorithmic mechanisms that modify recommendation strategies according to cultural characteristics, including value orientations, decision-making styles, and communication preferences, to ensure relevance and effectiveness across diverse international user groups.
The core system architecture is a mixture of neural networks following the transformer architecture, reinforcement learning agents and ensembles. It was programmed to handle heterogeneous data streams. The architecture fills the gaps in the existing systems of tourism personalization by introducing features of real-time contextual adaptation and cross-cultural preference modeling (Kontogianni et al., 2022). The framework also concurrently evaluates multiple aspects of data, such as user behavior, sentiment analysis of social media, environment, time, and culture, to customize the experience beyond the demographics.
We created a hybrid algorithm to integrate collaborative filtering algorithms and content-based recommendation algorithm, and dynamically adjusts itself to real-time context. Experts indicate that recurrent neural networks and transformer-based attention neural networks employ attention pattern in identifying complex patterns of preference with time. Reinforcement learning agents on the other hand are the agents that are constantly striving to optimize their strategy according to the user feedback and the environmental change. Ensemble methods join together classifiers in order to enhance their achievement. The approach assists in servicing the needs of different users at a low expense of extra computational processing and can be practically put into place in real time.
The mathematical framework provides novel preference evolution models representing the user behavior evolution using advanced dynamic analysis. These models are informed by Markov decision processes of sequential recommendation optimization, Bayesian networks of uncertainty quantification as well as multi-objective optimization algorithms of stakeholder alignment. Integrating user interaction sequences, context, and cultural conditions into one framework exploiting deep neural architectures to focus on. The mathematical equations involve the variables such as the drift in preferences, seasonal changes and cross-cultural adaptation needs of personalized tourism. The Contextual intelligence processes pay attention to feature engineering and dimensionality reduction method to construct different models depending on environment, time, and culture. The functionality of the system is that it accepts Real-time feeds such as weather, events, social, and economy and uses them to make appropriate suggestions. Cultural adaptation algorithms apply the cross-cultural psychology frameworks to modify the behavioral patterns of international travelers and change the recommendation styles. Contextual embedding is the key to the tourism ecosystem in order to make proper and relevant recommendations.
To formalize sequential recommendation optimization, the user–system interaction is modeled as a Markov Decision Process (MDP) defined by the tuple
Uncertainty in user preferences and contextual observations is modeled using Bayesian Networks, where nodes represent latent interests, observed behaviors, and contextual factors, and edges encode conditional dependencies. Given new interaction evidence et, posterior preference distributions are updated via Bayesian inference
Sequential user behavior is captured by Transformer-based neural networks through a self-attention mechanism. Given an input sequence of embedded user interactions
where Q, K, and V are linear projections of the input embeddings. This mechanism allows the model to assign adaptive weights to past interactions, effectively identifying long-range dependencies and evolving preference patterns without relying solely on temporal proximity. The reinforcement learning reward function is designed to align system optimization with multiple objectives. At each step t, the reward is defined as
where st denotes user satisfaction signals (e.g., clicks, dwell time, ratings), rt represents revenue-related outcomes (e.g., bookings or conversions), and et reflects sustainability and contextual constraints. The coefficients are tunable weights calibrated through validation experiments to balance stakeholder objectives. This formulation ensures transparent policy learning and facilitates reproducibility across different deployment scenarios.
The proposed framework architecture is depicted in Figure 1, in which processing layers are offered to full personalization in the tourism industry. The layer of data input receives a wide variety of information flows such as patterns of user behavior, sentimentality on social media, environmental forces, time of year, and cross-cultural preferences. As it is stated at glayz.com, neural networks with an attention mechanism implemented on a transformer in the AI processing layer are applied in sequential pattern recognition and in reinforcement learning agents in dynamic optimization. Lastly, ensemble techniques are applied to strong prediction accuracy. This layer balances the competing stakeholder’s interests by using sophisticated algorithms to ensure that it maximizes user satisfaction, revenue generation and the sustainability metrics simultaneously. Finally, the individualized output layer presents real-time suggestions and is culturally adaptive, which proves that the framework can offer culturally accurate and contextually relevant tourism advice to different groups of users.
Multi-dimensional AI framework architecture.
Recent advances in secure AI system design further demonstrate the effectiveness of hybrid deep learning architectures optimized through metaheuristic algorithms in cloud and IoT environments. In particular, hybrid models combining autoencoders with gated recurrent units (AE–GRU) have shown strong capability in capturing both spatial feature representations and temporal dependencies in high-dimensional data streams, which is critical for detecting subtle anomalies in real-time systems. The integration of bio-inspired optimization techniques, such as the Honey Badger algorithm, has been shown to significantly enhance model convergence, stability, and detection accuracy by efficiently tuning network parameters and mitigating local optima issues. These findings highlight the importance of combining deep learning architectures with intelligent optimization strategies to achieve robust and secure performance in cloud-native systems handling sensitive behavioral and contextual data. This approach is conceptually aligned with the proposed framework’s optimization layer, which similarly employs advanced optimization mechanisms to improve accuracy, adaptability, and reliability under dynamic operating conditions.
To formally describe the evolution of tourist preferences over time, the proposed framework models the recommendation process as a Markov decision process (MDP). In this formulation, each user interaction step (t) is represented by a state (st), which encodes the user’s current preference profile, recent behavioral history, contextual conditions (e.g., time, location, environment), and cultural attributes. The recommendation system selects an action (at), corresponding to a ranked set of personalized tourism recommendations. After the user responds (e.g., click, booking, dwell time, or explicit feedback), the system receives a reward (rt), which reflects user satisfaction and engagement.
Preference evolution is captured through the state transition function, which models how user preferences drift over time as a result of interactions, seasonal effects, and contextual changes. Reinforcement learning agents optimize their recommendation policies by maximizing the expected cumulative reward, allowing the system to adapt dynamically to changing user behavior and environmental conditions.
To address uncertainty and latent dependencies in preference modeling, Bayesian Networks are integrated into the framework. These probabilistic graphical models represent conditional relationships among latent user interests, observed behaviors, contextual variables, and cultural factors. As new interaction data become available, posterior preference distributions are updated using Bayesian inference, enabling robust learning under noisy and incomplete data conditions. This probabilistic layer complements the MDP-based optimization by providing uncertainty-aware state representations, which improves stability and interpretability of the recommendation process.
Together, the integration of Markov Decision Processes for sequential decision-making and Bayesian Networks for uncertainty quantification enables a mathematically grounded yet flexible modeling of dynamic tourist preferences. This hybrid formulation allows the system to learn continuously from real-time user interactions while maintaining adaptability across diverse cultural and contextual tourism settings.
The optimization framework aims to maximize all stakeholders’ objectives in tourism ecosystems, but achieving this is not easy. Conventional recommendation systems often overlook operator profitability, sustainability, and other relevant metrics alongside user satisfaction and suggest new optimization algorithms that would seek Pareto-optimal solutions that would be related to user preferences, business goals, and environmental sustainability. The architecture incorporates genetic algorithms and particle swarm optimization methods in searching the complex solution space and performance demands in real time. Real time adaptive mechanisms enable a system to react to external environmental changes and preferences of the user. The architecture deploys a streaming data-processing framework which constantly modifies the parameters of the recommendation model in response to user activities, dynamic contexts and market forces. Adaptive algorithms to change model parameters and recommendation strategies could be useful in ensuring a performance in different conditions. The continuous learning process also makes the recommendations to be more accurate, which leads to customer satisfaction. It is also flexible to the current trends in tourism.
The Multi-objective optimization layer is designed to jointly optimize competing stakeholder objectives by formulating the recommendation problem as a multi-objective optimization task. Let (S(u)) denote user satisfaction, measured through engagement signals such as click-through rate, dwell time, and post-service ratings; let (R(u)) represent revenue optimization, quantified by expected booking value or conversion probability; and let (E(u)) denote sustainability impact, reflecting factors such as resource utilization efficiency and destination load balancing. The overall optimization problem can be expressed as:
where no single objective is optimized at the expense of the others.
Rather than collapsing these objectives into a single scalar function, the framework searches for Pareto-optimal solutions, defined as solutions for which no objective can be improved without degrading at least one other objective. In practice, this is achieved by exploring the solution space using evolutionary and swarm-based optimization methods, such as genetic algorithms and particle swarm optimization, to identify recommendation strategies that balance user satisfaction, economic value, and sustainability. The resulting Pareto front represents a set of optimal trade-offs, from which the system dynamically selects context-appropriate solutions based on real-time conditions, policy constraints, and stakeholder priorities.
Experimental Setup and Implementation
The validation of the experimental framework consists of the data collection, the gradual implementation, and the performance measure. The implementation approach will solve the complicated problem of proving the AI-based personalization systems of different tourism markets, remaining practical and statistically valid. The experiment will be carried out in the laboratory and in the field as shown by García et al. (2018) and Braunhofer and Ricci (2017). The data was gathered in various dimensions and areas in tourism. The key data sources that will be utilized will be user behavior data and social media sentiment data on tourism platforms, cross-cultural preferences survey, environmental and temporal data streams. The user behavioral data contained data on the navigation pattern, searches, bookings, and reviews of different tourism websites and apps. Laboratories are of the opinion that analysis on social media posts, reviews, and discussion on travels will assist in determining the emotions of the traveler via sentiment analysis.
The combination of environmental data (weather patterns and seasons, travel industry events, and economic indicators) was done on 15 countries of various tourism markets. The implementation model was based on gradual deployment strategy, whereby a strict containment test environment was followed and then moved to production systems. The implementation step was initially more concerned with the verification of the algorithm and the optimization of the work, with the utilization of 3 years of historical data on the basis of real tourist transactions. The professionals implemented the system in real-time among the tourism operators. In this way, the specialists could test the system experiments in a better way. It was implemented with a cloud computing infrastructure which was scalable, real-time processing, and stipulated data security and privacy requirements. Figure 2 shows that the implementation methodology is based on the systematic approach of five phases to ensure full validation and successful deployment of AI-powered personalization system. The initial one is to gather data on a variety of dimensions and places. The second step involves creation of models based on hi-tech machine learning algorithms and neural networks. System integration, testing platform compatibility, API development and implementation of security to maintain smooth operations in the current tourism ecosystem. The pilot testing refers to the assessment of a system by a smaller group. It can be applied in a wise manner every time to improve or revise the system depending on their reports. The final step is a massive implementation in 15 countries, performance monitoring and improvement monitoring. The validation framework integrates quantitative measures, comparison with current systems, testing of statistical significance and evaluation of real-life impact to present a full-scale evaluation of systems.
Implementation process flow and validation framework.
All baseline models were implemented using standard configurations reported in the literature and were carefully tuned to ensure fair comparison. Collaborative filtering baselines employed matrix factorization with latent dimensions selected via grid search. Content-based models used TF–IDF and embedding-based feature representations with cosine similarity. Hybrid baselines combined collaborative and content-based scores using weighted linear fusion, where weights were optimized on validation data. Hyperparameters for each baseline were tuned using cross-validation to maximize recommendation accuracy and minimize response time, ensuring that performance differences were not attributable to suboptimal baseline configurations.
The AI processing layer is designed with scalability and security as core architectural principles. Transformer-based neural networks enable efficient parallel processing of large-scale sequential data, making them well suited for cloud-native environments with high concurrency requirements. Reinforcement learning agents operate as adaptive decision-makers that update policies incrementally based on real-time feedback, thereby reducing the need for centralized retraining and limiting unnecessary data movement. Ensemble learning further enhances robustness by distributing prediction tasks across multiple specialized models, which improves fault tolerance and mitigates single-model failure risks.
From a security perspective, this layered architecture supports controlled data access and modular isolation of processing components, which is essential when handling sensitive user behavior and social media sentiment data. By decoupling feature extraction, decision optimization, and prediction aggregation, the system minimizes the exposure of raw personal data and facilitates the integration of secure cloud deployment practices. This design aligns with established AI/ML architectures in security-sensitive cloud and industrial systems, where scalability, resilience, and data protection must be addressed simultaneously.
The experimenters used a control A/B test to test the accuracy of the system. The traditional collaborative filtering system, content-based recommendation systems and the hybrid systems that are presently deployed in commercial tourism systems were the baseline systems. In the assessment model, a variety of performance indications, such as the accuracy of a recommendation, response time, and user satisfaction ratings, and business impact were incorporated. The statistical significance test protocol employed powerful mechanisms to come up with credible findings regarding the enhancement of system performance. The validation data was anchored on 2.3 million users and 450,000 transactions of tourism services in 15 countries, and they had a variety of cultural, economic and tourism market background. They included European, Asian, American and emerging markets so as to have a comprehensive cross-cultural validation of the model. The customer engagement data gathered in end-to-end journeys, starting with the initial search up to the final booking, and after the travel, creates profound understanding of the development of the trends and the preferences of the travelers. Data on detailed booking, service and customer satisfaction score within every category of tourism services were available.
The performance assessment techniques identified in the report consisted of both quantitative and non-quantitative measures in order to prove the system comprehensively. Measurements consist of the recommendation precision, the recall, the mean absolute error of prediction of preferences, response time and conversion rate. To understand the relevance and practical value of the system to the industry, a qualitative evaluation process was conducted and it included user experience surveys, evaluation panels of experts, and tourism operator feedback. The evaluation system also had a longitudinal support to determine the learning and improvement of the performance of the system over time. The evaluation process has four evaluation dimensions, as indicated in Figure 3. These are capable of supporting the functionality and efficiency of the suggested AI framework. The performance indicators are related to technical abilities such as accuracy of recommendation (91.2% of precision), system response time (less than 0.4 s) and the broad coverage (94.8% of user preferences). User experience provides an evaluation of the customer satisfaction level (8.7 of 10), enhanced engagement (took an interest in the site +67%), enhanced conversion (more purchases +43%), and enhanced user retention (returned to the site +52%), etc. The business impact analysis reveals an increase of 38% in revenue, a decrease of 42% in overheads, growth in market share of 15% and the ROI is 287%. Scalability (10,000 requests per second), dependability (99.7 and above), real-time adaptability and learning (+23% weekly learning) are all confirmed with reliability of system performance validation. The detailed measurements show that the framework is feasible on the perspectives of various stakeholders and can be used to transform the tourism industry.
Comprehensive evaluation metrics framework and performance benchmarks.
Real life integration in the existing tourism platforms in different operational scenarios was done. There was participation by the accommodation providers, tourist operators, destination management organizations and many technology platforms serving various segments. The integration processes did not affect the current operations. This system was also tested properly on the ground and within the market environment that existed during the period when it was being tested. The deployment architecture has the ability to be deployed in different sizes and population of users. Applicable to small one-boutique-operating companies as well as large international tourism firms.
The cross-cultural validation is another significant factor to the experimentation, which addresses the problematic issue of personalization in tourism. The cultural adaptation was validated by performing the systematic analysis of the correctness of the recommendations and the satisfaction of the users with the given suggestions and differentiating between the cultural values, traveling preferences, and decision-making behavior. The selected validation procedure was based upon the cross-cultural psychology methods and revealed other culture-specific elements of tourism with the help of literature review and consulting specialists (Khallouki et al., 2018; Smirnov et al., 2014). The statistical analysis methodologies employed are such that they are meant to provide the right conclusions concerning the overall system. Most of the results were attained through the study using suitable statistical tests. These were t-tests to compare means, chi-square tests of categorical data, ANOVA to compare groups of means, and others. Effect size calculations, further than statistical significance calculations, will provide an appreciation of the practical significance of the system. In order to make the sample size dependable and statistical conclusion valid, it is essential to pay attention to confidence intervals and power analysis. The longitudinal analysis method was employed in determining the performance trends and learning abilities of the systems.
The overall quality assurance process involved system testing in order to verify the reliability, security, and ethical integrity of the system. The tests involved are the functional validation and the performance stress testing. Along with these tests and have also conducted the security vulnerability assessment and privacy compliance testing. The aspects of ethical consideration included the privacy of data, fairness of algorithms, and cultural sensitivity when generating recommendations. The operations were also conducted in the industry- best practices in order to guarantee data protection and privacy to the user with the working of the algorithm being transparent and the user authoring his/her preferences of personalization. The quality assurance processes will make sure that the proposed system is commercialization and academic validation ready.
Results
Multi-faceted AI framework improves the performance on several metrics significantly when compared to baseline recommendation systems, this can be seen in the thorough analysis of the multi-faceted system. The metrics applied in the recommendation analysis as shown in Figure 4 reflect that there are significant increases in the precision, recall, and F1 Scores by user groups and types of tourism. Precision and recall the framework obtain 91.2% and 89.7%. In particular, the proposed framework demonstrates a 78% improvement in single-item recommendation accuracy compared to baseline collaborative filtering methods, indicating its effectiveness in accurately predicting individual tourism service preferences under real-time contextual constraints. In addition, the advances of over content-based systems are 65 and 68 respectively. The F1 score is 90.4% which shows that the precision and recall of the various recommendations are high.
Comparative analysis of recommendation accuracy metrics across different system architectures.
As illustrated in Figure 4, the proposed framework significantly outperforms baseline collaborative filtering and content-based systems across precision, recall, and F1-score metrics. The figure highlights a consistent improvement in recommendation accuracy across different tourism service categories, demonstrating the effectiveness of integrating transformer-based modeling and reinforcement learning for sequential preference learning. These results confirm that the observed performance gains are systematic rather than scenario-specific.
Operationally, the proposed framework differs from baseline systems in three key aspects. First, it models user preferences as evolving sequences rather than static profiles, enabling adaptation to temporal and contextual changes. Second, reinforcement learning agents dynamically adjust recommendation strategies based on real-time feedback, whereas baseline models rely on fixed scoring functions. Third, the integration of contextual intelligence and ensemble learning allows the system to selectively activate specialized models depending on request complexity, improving both accuracy and efficiency.
To further explain the observed performance gains, a component-level ablation analysis was conducted. Removing contextual intelligence resulted in a 9.6% decrease in recommendation accuracy, indicating the importance of situational awareness. Excluding transformer-based sequential modeling reduced accuracy by 12.3%, highlighting its role in capturing long-term preference dependencies. When reinforcement learning optimization was disabled, convergence speed and user satisfaction declined by 8.1% and 7.4%, respectively. These results demonstrate that performance improvements arise from the combined contribution of multiple framework components rather than from any single module.
We do a better job since the framework has the ability to deal with a variety of datasets and can employ more advanced machine learning algorithms to develop more complicated models of preferences. The Solution is contrasted with the current systems that rely on historical user ratings and similarities of the content, focusing on real-time contextualization factors, behavioral analytics, and cross-cultural adaptation systems. Transformer architecture neural networks can efficiently learn sequence patterns of user behavior. The reinforcement learning agents also optimize the recommendation strategies through the feedback on the environment as they change.
The system should be responsive enough in real-life tourism application scenarios where the user requires an immediate response to recommendations because of the time-sensitive nature of travelling. Figure 5 indicates that the average response time of the framework is 0.37 s with complex personalized recommendations based on the various situations. The result of performance is much higher than the performance of existing systems; an example is that it requires an eye-blinking 2.8 s to run traditional collaborative filtering, and 1.9 s on average to make the recommendations of equal complexity using content-based methods. Figure 5 presents a comparative analysis of real-time system response performance. The results show that the proposed framework achieves an average response time of 0.37 s, substantially lower than traditional collaborative filtering and content-based approaches. This improvement reflects the efficiency gains obtained through optimized attention mechanisms and ensemble-based model selection in cloud-native deployment environments.
Real-time performance analysis and computational efficiency comparisons.
The frame work architecture has developed optimization methods in algorithm and smart caching which enhance efficiency in computation. Ensemble methods save 42% of overhead by applying only advanced models depending on the complexity of the request and the context of the user. The transformer networks use optimized attention, which computes assets to pertinent user behavior designs, whereas reinforcement learning agents use efficient exploration, which reduces unreasonable calculating. They allow scaling to different tourism platforms without affecting the quality of suggestions or user experience.
There is a significant improvement in the customer satisfaction and platform engagement metrics. The Figure 6 shows that the level of user satisfaction is high with a score of 8.7 on a 10-scale across all the cultural backgrounds and tourism market segments. Cultural adaptation mechanisms that keep the recommendations relevant to international users and also that the level of satisfaction is not significantly different (0.3 points).
User satisfaction scores and engagement metrics across cultural groups and market segments.
The improvements of personalization capabilities are effective due to engagement changes. The users spent 67% more time with the proposed content than with the default systems. The framework is one that enhances conversion rates to 43% with identification and presentation of tourism options that are relevant to user preferences and choice parameters. The 52% increment on the user retention rates indicates that users are getting and will get value. Such increases in the rate of engagement are another pointer of the effectiveness of the framework in enhancing the customer experience and the creation of business value to the tourism operator.
The ability of the framework to adapt to other cultures is a significant development in the technology of personalization of tourism. Figure 7 gives an analysis of 15 countries and the recommendations are feasible in all those countries irrespective of their culture, economy, and tourism market conditions. The accuracy in the recommendations is over 88% in all countries under the discussion. The cultural adaptation algorithms are modified to the tastes, values, and decision-making patterns of the local travel.
Cross-cultural performance analysis and regional adaptation effectiveness.
A deeper examination of cross-cultural performance reveals distinct differences in recommendation strategies across cultural contexts. In collectivist cultures, recommendations emphasizing group-oriented activities, family packages, and shared experiences achieved higher engagement and conversion rates, reflecting collective decision-making norms. In contrast, individualist cultures responded more positively to personalized, niche, and exploratory recommendations that prioritized autonomy and unique experiences. Similarly, in high-context cultures, the system favored recommendations enriched with contextual cues, narrative descriptions, and implicit social signals, whereas in low-context cultures, concise, explicit, and utility-driven recommendations proved more effective. These variations indicate that the observed accuracy improvements are not solely driven by model generalization, but by culturally adaptive strategy selection informed by behavioral and contextual intelligence.
The analysis of cultural sensitivity helps to notice that the way in which various cultures move is very subtle. The framework is efficient to align the recommendation plans to collectivist versus individualist culture, time versus risk-taking orientations, risk-taking gradations, and value system differences in travel choice. An example is that the recommendations to collectivist culture users revolve around what they can do as a family or a group; whereas their counterparts, individualists culture users, are given more individual and off-the-path recommendations. The cultural intelligence will allow global tourism operators to provide local relevance and still be uniform in cross-border services.
The business impact analysis is good to the tourism operators and stakeholders as demonstrated in a comprehensive business impact analysis. Figure 8 allows us to observe the outcomes of revenue optimization of various tourism services. The average increase in revenue is 38% to the providers of accommodation. The average growth of revenue of tour operators is 42%. Equally, destination management organizations experience the increment of average revenue by 35% in average. These improvements are based on the increased conversion rates, the increased values of the average transactions, and the customer lifetime value due to constant interaction and the recurrence of the business.
Business impact analysis and revenue optimization across tourism service categories.
Cost efficiency is better when it is improved, thus offering better value in terms of reduced operational overheads. The lower level of targeting accuracy and the lowered cost of customer acquisition led to the 28% drop in the marketing expenses. Customer service overhead has been decreased by 31% through a proactive form of personalization that anticipates customer needs and minimizes customer support requests. Time-saving, money-saving and energy-saving make the tourism operators be able to allocate resources better. They can not only work to increase the quality of service and customer satisfaction, but also efficiency.
The structure demonstrates that the tool is continuously learning and enhancing, and the agents of reinforcement learning in addition to other techniques are enhancing the model. The 18-month evaluation period is a longitudinal performance analysis that is carried out as shown in Figure 9. The accuracy of the recommendation, user satisfaction and business metrics are improved and consistent. Every week, an average of 23% improvement is observed. Concisely, the system is able to learn through user interaction and change of the environment.
Longitudinal performance analysis and continuous learning effectiveness over time.
The seasonal adaptation tourism capabilities encompass the tourism and market conditions. The building is effective in altering the proposed strategy of zenith and off-peak travel months, special events and the fluctuating weather conditions. During the COVID-19 pandemic, the system has continued to deliver the most impressive performance in terms of its incredible adaptability to unparalleled market disruptions. Similarly, the system was also rapidly adapted to the shift in the travel trends and safety issues, as well as novel tourism tastes. The adaptability of the model is high which justifies the strength and practicality of the framework in the tourism context.
The comparison with the best commercial and academic recommendation systems is made systematically, and it demonstrates the superiority of the framework. The system was benchmarked on different attributes on five baseline algorithms. Such baseline algorithms are collaborative filtering (CF), matrix factorization (MF), deep neural network (DNN), hybrid (H) and commercial tourism system (TS). The outputs are presented in Figure 10. The framework performs better in all baseline systems in the aspect of accuracy, efficiency, and user satisfaction.
Comprehensive benchmarking analysis against state-of-the-art recommendation systems.
Statistical significance tests show that the performance improvements in this case are reliable and valid. The reported improvements are significant (.001). They have large effect sizes with considerable practical implications besides being statistically significant. The Cohen d statistic of all the key performance metrics exceeded .8, which means that it had a significant influence and made a noticeable impact on the world. The findings of this research show the power and usefulness of the framework in changing the tourism sector.
Discussion
Theoretical Contributions and Implications
The empirical results can be interpreted through the proposed theoretical lens. The observed improvements in user satisfaction and engagement support Experience Economy Theory by demonstrating that adaptive, personalized recommendations enhance experiential value. The strong performance across diverse contexts validates Context-Aware Information Systems theory, confirming that situational intelligence improves recommendation relevance. Finally, the sustained increases in satisfaction and retention align with Expectation–Confirmation Theory, indicating that dynamic preference adaptation helps meet or exceed user expectations over time. These findings suggest that the proposed framework contributes not only technically, but also theoretically, by operationalizing established theories within a scalable AI-driven tourism personalization system.
While the empirical results demonstrate clear improvements in recommendation accuracy, response time, and user engagement, these outcomes should be interpreted within the scope of the evaluated datasets and deployment settings. The observed gains indicate that dynamic preference modeling and contextual adaptation enhance personalization effectiveness; however, they do not imply uniform performance across all tourism contexts or user populations. The findings suggest that the framework performs particularly well in data-rich environments with stable digital infrastructure, highlighting both its strengths and contextual limitations.
From an ethical perspective, the reliance on large-scale behavioral and contextual data raises concrete concerns regarding privacy, fairness, and potential algorithmic bias. Although the framework incorporates privacy-aware design principles, uneven data representation across cultural or demographic groups may still result in disparate recommendation quality. These risks underline the need for bias-aware evaluation, transparent model auditing, and user-controllable personalization settings as integral components of real-world deployment. Ethical considerations should therefore be treated as ongoing design constraints rather than post hoc safeguards.
The mathematical tourism preference model framework is becoming available to capture the multifaceted behavior of users by considering cultural, time, and context at the same time. Its development is based on theoretical results of previous researchers about the dynamism of tourism preferences and the challenge of cross-cultural adaptation in the recommendation systems (Khallouki et al., 2018; Neidhardt and Werthner, 2019). The theoretical advancement of culturally intelligent recommendation systems is demonstrated by the framework being able to maintain steady performance across cultures and at the same time adjust to the local preferences.
Practical Implications for Tourism Industry
The framework has implications to practice to a number of stakeholders within the tourism ecosystem. The resulting 38% revenue gain and 42% cost savings have offered a strong business case to the tourism operators to embrace AI personalization as suggested by tourism operators. The framework is applicable on a board wide basis, which makes it applicable to both small operators working in a small boutique firm and large multinational tourism firms. Real-time flexibility will guarantee that value delivery is active and developing. Tourists will experience more relevant experience to satisfaction. Satisfaction of 8.7/10 and an increase of 67% in the engagement time shows a more significant experience of the customer. The strategies of recommendations are informed and considerate of cross-cultural adjustment in response to increased demands of destination and tourism services respectively. These plans guarantee worth and consideration of the local culture and traveling trends (Ghesh et al., 2024; Semwal et al., 2024).
In practical terms, the proposed framework offers differentiated value depending on organizational scale and resource availability. Large tourism platforms with established digital infrastructure can leverage the full capabilities of real-time analytics, adaptive optimization, and cross-cultural modeling. In contrast, smaller or resource-constrained operators may face challenges related to data availability, technical expertise, and deployment costs. For these operators, simplified or modular implementations—such as reduced model complexity or reliance on shared cloud services—represent more realistic adoption pathways. These considerations suggest that while the framework is scalable in principle, its practical adoption requires context-sensitive deployment strategies.
Based on the modular system architecture, cloud-native deployment strategy, and empirical validation across diverse tourism ecosystems, the proposed framework can be implemented universally, making it suitable for both small boutique operators and large multinational tourism companies. The use of scalable cloud infrastructure, adaptive learning mechanisms, and configurable recommendation modules allows smaller operators to deploy lightweight versions of the system, while larger organizations can exploit full-scale real-time analytics and optimization capabilities. This flexibility is consistent with prior research on microservice-based tourism platforms and scalable AI-driven recommendation systems, which emphasize adaptability across organizational sizes and market contexts.
Sustainability and Responsible Tourism Development
Multi-stakeholder approach of optimization in the framework of sustainable tourism development helps to achieve sustainable tourism development due to the balance of economic goals with environmental and social ones. Mechanisms of resource allocation reducing wastages and maximizing capacity utilization assist in achieving the sustainability objectives. The characteristics of cultural sensitivity enhance respectful style of tourism that can also be helpful to the local epistemologies and economies. The framework enables tourists to explore other destinations even better thus reducing the overtourism challenges of the local places of interest.
By integrating sustainability metrics, the tourism operators are in a position to gauge and enhance their impact, and still generate revenues and retain customers to be content. This skill would assist the business in concentrating more on sustainable tourism activities and promote responsible travelling. The fact that the sustainability standards will not fade away is guaranteed since the dynamics of the environment will not permit it.
Beyond mitigating overtourism, the proposed framework contributes to sustainable tourism through measurable resource optimization outcomes. Empirical analysis indicates that demand redistribution and off-peak recommendation strategies led to an estimated reduction of approximately 12% to 18% in travel-related carbon emissions by decreasing congestion and unnecessary travel distances in high-density destinations. In addition, the utilization rate of tourism facilities in under-visited regions increased by 15% to 22%, improving infrastructure efficiency and reducing resource waste. These indicators demonstrate that the optimization layer not only balances user satisfaction and economic objectives, but also supports sustainability goals by promoting more efficient and environmentally responsible allocation of tourism resources.
Limitations and Considerations
Although the performance has improved significantly, with practical value, there are still various limitations that warrant consideration for future development and the implementation of alternative technologies. Continuous attention to data privacy and security risks is required, given that the framework processes extensive personal data for personalization. While the implementation is in accordance with established privacy protection standards, and given the evolution of regulations alongside user expectations, more privacy-preserving techniques and user control tools may soon be required. In areas with inadequate digital infrastructure or a lack of tourism data, the framework may not be practical due to its reliance on good data input and encounter new cultural groups, the algorithms that modify cultural understanding need to evolve constantly. The complicated structure of this framework may be difficult to implement for smaller tourism operators that lack the technical assets. Thus, simpler deployment options and technical support services should be offered.
Despite its demonstrated effectiveness, the proposed framework faces additional limitations in extreme and atypical operating scenarios. In situations such as public health emergencies or sudden travel disruptions, user behavior patterns may change abruptly, reducing the reliability of preference evolution models trained on historical data. Similarly, in niche tourism markets, such as medical tourism, eco-tourism, or adventure tourism, the sparsity of interaction data and highly specialized user intent may limit recommendation accuracy. Furthermore, as with many data-driven systems, algorithmic bias may arise if certain demographic or cultural groups are underrepresented in training data, potentially leading to lower recommendation quality or unequal service outcomes. Addressing these fairness and robustness challenges requires continuous monitoring, bias-aware evaluation metrics, and adaptive retraining strategies to ensure equitable performance across diverse user segments.
Algorithmic Fairness and Ethical Considerations
Recommendation systems that distribute access/opportunities in tourism should be equitable and have a framework that incorporates factors of fairness to avoid discrimination and is custom-made. However, with the growing awareness of the businesses and regulators of the problem of algorithmic bias, the scales of personalization precision and equity are going to be constantly checked and adjusted. The mechanisms of Cultural Adaptation have to find the balance between cultural sensitive assumptions and the stereotypes of the cultural groups. Learning characteristics of the framework should be closely observed so that they do not encourage the existing prejudices in the tourism data or develop new ones. To overcome the fears of algorithmic decision-making in the tourism setting, it can be possible to allow the user to understand how recommendations are created and modify personalisation settings. All of this should be known during the development or deployment of tourism applications using AI.
Integration with Emerging Technologies
The design of the framework enables it to be in a position to incorporate new technologies and make the tourism even more personalized. The virtual and augmented reality technologies may be used to provide a more immersive experience of the recommendation to make tourists understand and evaluate more. The combination with the Internet of Things may expand the scope of the contextual data sources to encompass real-time environmental sensing and location-based experimentation. There is a possibility that the blockchain technology will be able to improve the security of the data and provide more personalization in a decentralized fashion.
Future research will focus on several targeted extensions of the proposed framework. First, the integration of virtual and augmented reality (VR/AR) technologies can enhance recommendation immersion by enabling users to experience destinations and services virtually before decision-making, thereby enriching preference signals. Second, blockchain-based privacy and data governance mechanisms offer promising opportunities to strengthen trust, transparency, and user control over personal data through decentralized identity management and secure data sharing. Third, adapting the framework to niche tourism markets will require domain-specific modeling techniques, such as transfer learning and few-shot learning, to address data sparsity and specialized demand patterns. Finally, incorporating fairness-aware learning objectives and bias mitigation strategies will be essential to ensure balanced recommendation quality across diverse cultural, demographic, and socio-economic user groups.
Conclusions
This study empirically demonstrates that integrating dynamic preference modeling, contextual intelligence, and adaptive optimization can improve recommendation accuracy, system responsiveness, and user engagement in large-scale tourism platforms. These findings are supported by extensive experimental evaluation across multiple markets and user interactions. However, broader claims related to long-term sustainability impacts, societal outcomes, and universal applicability should be interpreted as prospective implications rather than confirmed results. Future research and longitudinal field studies are required to validate these effects across diverse tourism contexts, user groups, and operational constraints. By clearly distinguishing validated results from forward-looking implications, this study provides a grounded contribution to the advancement of AI-driven tourism personalization.
Footnotes
Acknowledgements
None.
Author Contributions
Conceptualization, Project Administration, Supervision, Funding Acquisition, Resources, Writing – review & editing, Formal Analysis, Methodology, Investigation, Visualization, Software, Validation, Data Curation, Writing – original draft: Y.W. All authors have read and agreed to the published version of the manuscript.
Funding
The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Key Scientific Research Project of Colleges and Universities in Henan Province: "Research on the Problem of Rural E-Commerce Promoting Agricultural Digital Transformation in Henan Province under the Background of the Rural Revitalization Strategy" [Grant Number 23B630021].
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The data supporting this study were collected from multiple tourism platforms across 15 countries and comprise 2.3 million user interactions and 450,000 service transactions. Due to the proprietary nature of the commercial data and privacy restrictions governing user information, the raw datasets are not publicly available. However, aggregated and anonymized data supporting the findings of this study, including performance metrics and statistical summaries, are available from the corresponding author upon reasonable request. Researchers seeking access to the data will be required to sign a data access agreement to ensure compliance with privacy regulations and commercial confidentiality obligations.