Explainable AI-Assisted Theory Generation and Mechanism Identification for Sustainable Tourism: A Methodological Framework

摘要

As research and practice in sustainable tourism continue to evolve, the field increasingly faces a mismatch between rapidly changing practical challenges and the slower pace of theoretical development. Much of the existing literature relies on traditional qualitative approaches—such as Grounded Theory—or descriptive quantitative analysis. While valuable, these methods often struggle to systematically generate new theoretical insights from large-scale, multi-source datasets. Their limitations become particularly evident when attempting to capture the dynamic, nonlinear, and highly interconnected nature of tourism systems, where multiple social, environmental, and behavioral factors interact in complex ways. To address these challenges, this paper proposes a methodological framework that integrates Explainable Artificial Intelligence (XAI) to support theory generation and mechanism identification in sustainable tourism research. The framework leverages widely accessible data sources and cost-effective analytical tools, combining the strong pattern-recognition capacity of machine learning models with causal inference techniques. Through this integration, it establishes a structured pathway that moves from data-driven discovery to theoretically meaningful explanation, enabling researchers to uncover underlying mechanisms rather than merely identifying correlations. To demonstrate the practical application of the framework, the paper uses the formation mechanism of tourists’ environmentally responsible behavior as an illustrative case. The analysis integrates multi-source heterogeneous data, including tourist reviews, social media activity, and geospatial information. Within this framework, XAI methods such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) are employed to interpret the internal decision logic of complex predictive models, including deep learning networks. In addition, causal discovery algorithms, such as the PC algorithm and the Fast Causal Inference (FCI) algorithm, are applied to explore potential causal relationships among the identified variables. The results show that XAI not only improves the transparency and interpretability of predictive models, but also plays a crucial role in revealing the key driving factors and their interactive relationships underlying sustainable tourism behaviors. By identifying how different variables influence outcomes and how they interact with one another, the framework helps explain the “why” and “how” behind observed tourism phenomena. This process enables researchers to move beyond simple correlation analysis and toward the development of theoretical constructs and propositions with causal significance, thereby facilitating the emergence of new theoretical insights. Overall, the methodological framework proposed in this study offers a new research paradigm for sustainable tourism studies. By bridging the gap between the predictive strength of big-data analytics and the explanatory depth required for theory development, the framework accelerates the process of theoretical innovation in the field. At the same time, it provides tourism destination managers and policymakers with more precise, evidence-based, and forward-looking decision-support tools, ultimately contributing to the advancement of sustainable tourism development.