The low-carbon advancement of ports is essential for global carbon reduction. A scientific evaluation of ports' CO2 emission efficiency (CEE) is a crucial prerequisite for facilitating their low-carbon transition. Nevertheless, current research typically neglects the heterogeneous characteristics of port transportation services. The data envelope analysis (DEA) model for non-homogeneous decision-making units (DMUs) can effectively evaluate these DMUs, but existing models neglect the handling of undesirable outputs. This study addresses the shortcomings of the DEA model for non-homogeneous DMUs by using a weak disposability approach to handle undesirable outputs and Stackelberg game theory to solve input-output split ratios. The proposed model is utilized to evaluate the CEE of 88 Chinese port transportation services from 2018 to 2021. The results indicate that (1) the average overall CEE of Chinese ports remains at a relatively low level, indicating significant potential for improvement. (2) From 2018 to 2021, the overall average CEE continued to increase, though the growth rate showed a tendency to decelerate. This suggests that the series of low-carbon policies implemented by the government have yielded positive results, yet their effectiveness has been partially undermined by global emergency events. (3) Container transshipment, being the most standardized process, exhibits the highest average CEE, reflecting the contribution of smart port initiatives to low-carbon development. (4) Regression analysis reveals that port infrastructure scale and foreign trade value are primary drivers of CEE, whereas carbon reduction policy pressure and regional economic development level exhibit suppressive impacts. The case study validated the effectiveness of the proposed method and provided a reference for the evaluation of non-homogeneous DMUs with undesirable outputs.

Port activities have become deeply embedded in the development trajectories of port cities, and their impacts on urban environmental pressure cannot be characterized as a simple linear scale effect. Instead, these impacts exhibit pronounced stage-dependent and structure-contingent features. Focusing on major port cities in China, this study develops an integrated analytical framework that combines fixed-effects residualization, interpretable machine learning, and city typology to systematically characterize the nonlinear relationships among port activity intensity, transportation and freight structure, and urban environmental pressure. Specifically, the study first employs an Extreme Gradient Boosting (XGBoost) model coupled with Shapley Additive Explanations (SHAP) decomposition to identify endogenous threshold effects of port activity intensity on multiple categories of environmental pollution intensity, without imposing any a priori functional form. This approach reveals key changes in the marginal environmental effects of port activities across different stages of development. Subsequently, a Gaussian mixture model (GMM) is applied to classify port cities into structurally distinct types, capturing long-term differences in port dependence, transportation organization patterns, and baseline pollution conditions. Building on this typology, the study conducts type-wise SHAP decomposition and pre-/post-threshold comparative analysis to trace the underlying mechanisms through which thresholds emerge across different categories of port cities. This analysis interprets transportation structure and existing pollution baselines as potential amplifying or buffering pathways in the model-implied contribution structure, depending on city type. The results demonstrate substantial heterogeneity in both the location and environmental implications of port activity thresholds across port city types, indicating that environmental risks are not determined solely by port scale but are contingent on the urban structural conditions. By emphasizing stage-specific dynamics and structural heterogeneity, this study advances the understanding of environmental effects in port cities and provides empirical evidence to inform differentiated port development regulation, transportation system optimization, and targeted environmental governance strategies.

Climate anxiety has emerged as a significant global psychological and social response to climate change, potentially shaping public engagement and support for climate-related technologies and policies. Here we develop a framework for analyzing online climate anxiety using social media data from China based on 177,232 geo-referenced Weibo posts from 2010 to 2024. The analysis began with the investigation of climate anxiety themes using climate-anxious dictionaries and machine learning methods. Next, the emotional intensity of climate anxiety was assessed through the semantic similarity-based scoring approach. Finally, statistical models were applied to measure the factors influencing climate anxiety. Four major findings are arrived. First, extreme weather events (52.36 %) and livelihood and resource insecurity (22.87 %) were the most discussed and concerning themes, with a notable increase in discussions during summer and autumn. Second, the intensity of climate anxiety has risen significantly. The average intensity increased from 4.42 during the period of 2010-2017 to 7.08 during 2018-2024, with a further notable rise to 7.49 in the more recent period from 2020 to 2024. Third, regions such as Beijing (8.70), Guangdong (8.31), and Zhejiang (7.94) exhibited the highest levels of climate anxiety. Fourth, the intensity of climate anxiety is associated with key demographic and regional factors. Specifically, younger individuals and those residing in climate-vulnerable or informationally developed regions exhibited stronger emotional responses. The framework provides a scalable method for tracking the spatiotemporal dynamics of collective climate anxiety online. The findings demonstrate that digital expressions of climate anxiety constitute a measurable indicator of public concern and carry significant implications for anticipating societal responses and designing targeted communication within climate governance.

The New International Land-Sea Trade Corridor is located in western China. It connects the Silk Road Economic Belt, the 21st Century Maritime Silk Road, and the Yangtze River Economic Belt, and holds an important strategic position in the regional coordinated development. Aiming at the high-dimensional decision-making problem of multimodal transport path planning for the new corridor, this paper proposes a path planning algorithm that combines a multi-hierarchy heterogeneous graph attention network with deep Q-network (MHGAN-DQN). The algorithm dynamically incorporates the multiple heterogeneous relationships between nodes and edges in the transportation network into the action selection process, significantly enhancing the model's ability to capture key paths. Additionally, the algorithm introduces a node decomposition method to ensure consistency in the reward of action selection, avoiding conflicts with the requirements of the Bellman equation, effectively simplifying the state space management in path planning. Finally, this paper thoroughly examines the multi-objective impact of cost, time, and carbon emission on multimodal transport path selection. It further explores the comprehensive impact of railway freight subsidies, carbon tax policies, and uncertainties in transportation time on the final transportation scheme. The research results significantly improve the accuracy and efficiency of path decision-making, deepen the application of deep reinforcement learning in multimodal transport, and provide decision-making support for policy development related to the sustainable development of logistics systems of the New International Land-Sea Trade Corridor.

Purpose Cruise tourism live streaming has become an integral part of cruise companies’ marketing strategies; however, the mechanism that links cruise tourism live streaming interaction and consumer purchase intention remains underexplored. Therefore, this study aims to examine the impact of cruise tourism live streaming interaction on consumer purchase intention as well as the mediating roles of perceived trust and perceived value within the Stimulus–Organism–Response (SOR) model. Design/methodology/approach A survey was conducted to test the research model, and data from 456 respondents were analyzed using SPSS 27.0 and AMOS 28.0. Findings The results show that cruise tourism live streaming interaction (content interaction, form interaction and interpersonal network interaction) significantly affects consumer purchase intention, and perceived trust and perceived value mediate this relationship. The findings further support the applicability of the SOR model in the context of cruise live streaming. Practical implications This study offers actionable guidance for cruise operators to optimize their live streaming strategies. By enhancing content quality, interaction efficiency and community building, companies can effectively foster viewer trust, amplify the perceived value of cruise products and, ultimately, stimulate cruise booking conversions. Originality/value This study enriches the literature on tourism live streaming by conceptualizing cruise tourism live streaming interaction across three dimensions: content interaction, form interaction and interpersonal network interaction. In contrast to previous research, this study applies the SOR model to cruise tourism live streaming, demonstrating how interaction stimulates purchase intention through the chain mediators of perceived trust and perceived value. Additionally, this study provides practical insights for cruise corporations on optimizing live streaming interaction within their marketing strategies to drive cruise bookings.

Artificial Intelligence (AI), with its advanced data analysis and decision-making capabilities, provides new opportunities for improving transportation efficiency and promoting sustainable development. This paper presents a systematic review of recent advances in AI applications for green transportation, focusing on five key areas: intelligent scheduling and traffic optimization, unmanned technologies, green transportation infrastructure, shared mobility services, and policy support. The review highlights that AI technologies can effectively improve traffic management, optimize resource allocation, reduce energy consumption, and decrease carbon emissions through intelligent decision-making and data-driven approaches. Furthermore, AI-enabled autonomous systems, smart infrastructure, and shared mobility services contribute to the development of safer, more efficient, and low-carbon transportation systems. However, several challenges remain, including data privacy, system integration, model robustness, and policy coordination. Future research should focus on strengthening interdisciplinary collaboration, integrating emerging technologies, and developing effective governance frameworks to support large-scale deployment. This review provides comprehensive insights into the current progress, challenges, and future directions of AI-driven green transportation, offering valuable guidance for researchers and policymakers.

Existing policy evaluation frameworks often fail to capture multi-dimensional policy characteristics, particularly in complex governance systems. To address this, this study develops a novel quantitative decoding framework that integrates text mining techniques with the Policy Modeling Consistency Index (PMC‑Index) model to enable objective, multi-dimensional policy assessment, thereby reducing the subjectivity inherent in traditional variable selection. Applying this framework to 25 representative national‑level green shipping governance policies in China from 2011 to 2025, this study finds that the policies exhibit an excellent overall consistency. However, the framework also identifies three structural weaknesses: insufficient cross‑departmental collaboration, lack of clear short‑/medium‑/long‑term implementation timelines, and over‑reliance on demand‑side regulatory tools. Based on these findings, this study proposes targeted optimization pathways including multi‑level collaborative governance, balanced policy instrument mixes, and integration into global carbon pricing frameworks. This framework solves the subjectivity problem of variable selection in traditional PMC-Index model and realizes the objective decoding and multi-dimensional quantitative evaluation of complex policy texts, which provides a new scientific method for quantitative evaluation of cross sectoral and multi-dimensional policies.

Sustainable agricultural systems are crucial for balancing food security and ecological protection. This study develops a two-stage inverse network data envelopment analysis (DEA) model that incorporates shared inputs and undesirable outputs to evaluate and optimize resource allocation in agricultural production and pollution control. Using data from 31 Chinese provinces (2010-2023), the model estimates optimal resource allocation strategies under constant-efficiency and efficiency-improvement scenarios. Results indicate that although system efficiency is generally improving, notable regional disparities remain. Under constant efficiency, achieving a 5 % output increase requires substantial input growth, particularly in pesticides, whereas efficiency improvement reduces overall inputs by an average of 5.84 %, indicating the role of technological progress in resource conservation. The proposed framework represents a dynamic and practical tool for policymakers to design targeted, forward-looking strategies for sustainable agriculture.

Ship emissions constitute a major source of air pollution in coastal regions. To address the challenges associated with high-precision and dynamic emission monitoring, this paper proposes a cooperative unmanned aerial vehicle (UAV) monitoring framework that integrates data envelopment analysis (DEA) with multi-agent reinforcement learning. Specifically, a path planning and sampling decision-making model based on multi-agent proximal policy optimization (MAPPO) is developed, and a DEA-based cross efficiency social reward (CESR) mechanism is introduced to characterize the relative efficiency and cooperative relationships among UAVs under multiple objectives, including energy consumption, spatial coverage, and emission inversion performance. This mechanism guides UAVs to form an appropriate division of labor in dynamic, multi-source, and uncertain environments, thereby reducing redundant sampling and improving overall monitoring efficiency. Furthermore, a unified simulation environment is established by coupling a pollution plume dispersion model driven by realistic wind fields, a ship emission inversion model, and UAV dynamics and energy consumption constraints. Based on real automatic identification system (AIS) ship trajectories and ERA5 wind field data from the offshore area of the Copenhagen Port, Denmark, systematic comparative experiments are conducted under scenarios with varying ship densities and wind field complexities. The results demonstrate that the proposed method achieves significant improvements in emission inversion accuracy, detection capability, energy efficiency, and cooperative stability. In particular, it exhibits stronger robustness under conditions of high ship density and intense wind field disturbances. These findings provide a new cooperative decision-making paradigm for multi-UAV ship emission monitoring that combines performance enhancement with improved interpretability.

China's transport sector is a major contributor to energy use and carbon emissions, requiring a low-carbon transition. Emerging transport technologies, including electrification, hydrogen, and intelligent systems, are modeled as technological progress improving energy efficiency. This study develops a multi-sector computable general equilibrium (CGE) model, disaggregating transport into rail, road, water, air, and other subsectors, to evaluate multiple carbon tax rates and revenue recycling scenarios. The analysis examines the synergistic effects of carbon taxation, revenue recycling, and technological progress on emissions, economic growth, and energy structure. Results show that carbon taxation without revenue recycling effectively reduces emissions but leads to GDP losses, achieving an environmental dividend without an economic efficiency dividend. When combined with revenue recycling, economic losses are mitigated, though emission reductions are partially offset. When further combined with technological progress representing emerging transport technologies, carbon taxes amplify emission reductions while alleviating negative impacts on sectoral output and the broader economy, and under certain scenarios a modest GDP gain is observed, suggesting the possibility of coordinated environmental and economic benefits. These findings provide quantitative evidence to guide carbon tax design and support strategic adoption of low-carbon transport technologies in China.

Cross-border e-commerce transaction fraud has emerged as a focal point of current concern, and the application of machine learning algorithms for fraud detection is becoming increasingly prevalent. However, existing methods lack a universal integrated analytical framework, have limited applicability, and are inadequate in recognizing identity concealment. To address these challenges, a generative dynamic transaction network aggregation analytical framework is proposed. This framework first achieves dynamic classification of transactions through similarity-weighted sums of nodes with the same attributes and adaptive weighting. Next, a backward merging mechanism is introduced to construct an associated transaction network that captures latent behavioral connections. Additionally, a Generative Adversarial Network is utilized to enhance fraudulent samples, thereby alleviating class imbalance. An improved hybrid gray wolf optimization algorithm is employed for global optimization of weights, enhancing the model's adaptability. Empirical research conducted on a real cross-border transaction dataset demonstrates that the proposed framework significantly outperforms traditional aggregation feature methods in terms of accuracy, precision, recall, and F1 score (p<0.05). Furthermore, the core mechanisms of the framework demonstrate general applicability, as they do not rely on preset features of specific datasets; it can be utilized with any data that includes identity and transaction information. This framework is extendable to fraud detection scenarios involving identity concealment and transaction behavior analysis, such as in cross-border e-commerce and online payments.