This paper develops a graph-based dynamic opinion-formation framework for group decision-making that addresses a limitation of conventional consensus-oriented models: in many real decision environments, full agreement among experts is neither realistic nor required. Existing approaches are designed to force convergence to a single consensus and therefore cannot represent settings in which persistent disagreement naturally emerges while a collective decision still needs to be made. To fill this gap, we propose a dynamical system in which experts revise their opinions through heterogeneous interpersonal influences on a general graph and individual self-trust. The resulting dynamics admit two stable and internally coherent opinion groups, enabling collective decisions to be derived from the emergent polarized structure rather than from enforced consensus. We analytically characterize the long-term behavior of the model under general network structures and illustrate its decision-making implications through simulation studies. The results show how a heterogeneous graph network shapes the formation of opinion groups and the associated collective decision. The framework thus offers a methodological tool for graph-based group decision processes in which stable disagreement, rather than full consensus, is the expected outcome. In addition, its intrinsic bipolar structure makes the framework particularly effective for identifying a single best alternative by naturally amplifying the separation between the best option and others.

Modeling the collective dynamics of complex systems, represented as graph-structured time series, is a central challenge in statistical physics and applied science. Prevailing Spatio-Temporal Graph Neural Networks (STGNNs) are constrained by a serial processing paradigm that creates structural information bottlenecks, leading to the premature abstraction and loss of fine-grained correlations between microscopic system states. This paper posits a hierarchical design principle: the efficacy of adaptive mechanisms, such as attention, is fundamentally contingent upon the integrity of the underlying feature propagation architecture. To validate this, we introduce DMA-EISTGCN, a framework featuring two innovations: (1) an Early Spatio-Temporal Interaction (EI) module, a non-serial design that ensures lossless feature fusion at the model’s front-end, and (2) a Dynamic Multi-scale Attention (DMA) mechanism that adaptively arbitrates between channel, spatial, and regional feature refinement. Experiments on four real-world traffic forecasting benchmarks show the model establishes a new state-of-the-art, reducing prediction error by up to 14.0% compared to strong SOTA baselines. Critically, an ablation study reveals that applying the adaptive attention mechanism to a conventional serial backbone degrades performance. This provides direct empirical validation for our central thesis. This work establishes a new design paradigm for spatio-temporal models, demonstrating that resolving architectural bottlenecks is a necessary precursor to unlocking the full potential of adaptive learning in complex forecasting tasks.

The continuous growth in aviation traffic demand has made airport slot allocation increasingly challenging. Given the complexity of the problem, reinforcement learning methods have gradually emerged as an effective approach to solving it. This paper proposes a deep reinforcement learning method to address the airport slot allocation problem, utilizing a Sequence-to-Sequence structure policy network for training to learn slot allocation strategies. We first formulate the airport slot allocation problem as a Markov decision process using a decomposition method along the time axis. Then, we design an encoder to efficiently capture state information and employ a heuristic masking generation method in the decoder to ensure the feasibility of the solution to the airport slot allocation problem, thereby providing high-quality solutions. Evaluations on real-world airport slot allocation problems show that our DRL method performs well compared to traditional heuristics and other DRL methods. Additionally, the trained policy network exhibits good generalization capabilities across instances of different sizes.

The rapid electrification of urban transportation constitutes a significant phase transition in modern energy systems, yet it faces a critical bottleneck: the optimal deployment of charging infrastructure. This facility location problem features high-dimensional, non-convex landscapes and stochastic demand, rendering deterministic approaches inadequate. Existing planning methods relying on precise time-series forecasting suffer from error propagation and premature convergence in complex urban topologies. This study proposes a novel Demand Modeling framework that estimates the relative “demand potential” of Traffic Analysis Zones (TAZs) by integrating historical trajectories with spatial semantic features to identify intrinsic demand attractor. To solve the resulting location-allocation problem, we introduce the Chaos-Enhanced Adaptive Memetic Algorithm (CE-AMA). This method synthesizes evolutionary global search, ergodic exploration via a hybrid Logistic-Tent chaotic map, and memetic hill-climbing local exploitation. Theoretical analysis maps the optimization objective to a system Hamiltonian, showing that chaotic perturbations couple with thermodynamic temperature to induce a “reheating” effect, allowing the system to escape metastable glassy states. Validated against the Shenzhen UrbanEV dataset under a strict 14,400-evaluation budget, CE-AMA achieves a superior scalarized optimum of 0.2226, outperforming Simulated Annealing (0.3541), Particle Swarm Optimization (0.5169) and Genetic Algorithms (0.2804). The optimized configuration delivers 62.1% coverage and an 86.9% service level with robust 94.4% cost efficiency. These results demonstrate that introducing mathematically grounded controlled disorder via chaos theory effectively navigates the rugged energy landscape of urban infrastructure planning, minimizing system configurational entropy.

Baltimore's transportation system mainly relies on walking and buses, with researchers aiming to improve these by optimizing bus routes and sidewalk facilities. However, costs and traffic congestion remain concerns, especially after the collapse of the Francis Scott Key Bridge, which exacerbated congestion and affected travel time, freight costs, and supply chains. For Task1: To address these issues, researchers built a hypernetwork model for a multimodal transportation system considering transfer relationships. This model analyzes traveler behavior, defines feasible paths, and establishes cost functions, providing a comprehensive perspective for decision makers to identify bottlenecks, optimize resources, and promote sustainable development during bridge reconstruction, ensuring smoother transitions and minimizing disruptions. For Task2: The researchers conducted a cost analysis, studying travel time, emissions, and energy consumption for different modes of transportation. They employed a multimodal transportation network flow allocation model and sensitivity analysis to solve programming problems, aiding in quantifying the project's impact on stakeholders and providing a robust scientific basis for decision making. For Task3: Based on these analyses, the researchers proposed various measures to improve the transportation network, including setting up low-emission zones, promoting new travel modes, and reducing private car entry. They also suggested supportive measures such as strengthening parking facilities, implementing differentiated fees, optimizing traffic layout, and integrating public and shared travel options. Additionally, they proposed long-term development plans, policy guarantees, and encouraged public participation to create a green and efficient travel environment. Overall, this research report comprehensively evaluates the potential impact of optimizing bus routes and sidewalk facilities in Baltimore, proposing a range of measures and long-term plans to foster an efficient, environmentally friendly, and sustainable transportation system that caters to the city's current and future needs.

In the financial sector, credit risk represents a critical issue, and accurate prediction is essential for mitigating financial risk and ensuring economic stability. Although artificial intelligence methods can achieve satisfactory accuracy, explaining their predictive results poses a significant challenge, thereby prompting research on interpretability. Current research primarily focuses on individual interpretability methods and seldom investigates the combined application of multiple approaches. To address the limitations of existing research, this study proposes a two-stage interpretability model that integrates SHAP and counterfactual explanations. In the first stage, SHAP is employed to analyze feature importance, categorizing features into subsets according to their positive or negative impact on predicted outcomes. In the second stage, a genetic algorithm generates counterfactual explanations by considering feature importance and applying perturbations in various directions based on predefined subsets, thereby accurately identifying counterfactual samples that can modify predicted outcomes. We conducted experiments on the German credit datasets, HMEQ datasets, and the Taiwan Default of Credit Card Clients dataset using SVM, XGB, MLP, and LSTM as base classifiers, respectively. The experimental results indicate that the frequency of feature changes in the counterfactual explanations generated closely aligns with the feature importance derived from the SHAP method. Under the evaluation metrics of effectiveness and sparsity, the performance demonstrates improvements over both basic counterfactual explanation methods and prototype-based counterfactuals. Furthermore, this study offers recommendations based on features derived from SHAP analysis results and counterfactual explanations to reduce the risk of classification as a default.

To address the issues of low accuracy and high misclassification rates in financial distress prediction (FDP) models, this study proposes a classifier model based on a backpropagation algorithm and a discriminant-restricted Boltzmann machine (BP-DRBM). Furthermore, a two-stage selective ensemble approach, employing stepwise selection and majority voting methods, is implemented to enhance model performance. An empirical analysis is conducted using data from 1,440 Chinese listed companies from 2018 to 2023. The results indicate that: (1) the BP-DRBM model achieves an accuracy of 85.7

Environmental policies are a primary tool for governments to address environmental issues, yet their effectiveness remains debated. This study examines the direct and indirect effects of regional environmental policies on environmental efficiency. Based on the environmental policies collected from 150 official websites, the environmental policy intensity (EPI) and convergence of 30 Chinese regions are quantified by text mining methods, and the Spatial Durbin Model is employed to investigate the direct and spillover effects of environmental policies on environmental performance. The results indicate that regional environmental policy convergence (EPC) has statistically significant and positive direct and spillover effects. Conversely, stricter EPI tends to hinder sustainability in both local and neighboring regions. Furthermore, empirical analysis demonstrates that EPC generates significantly greater positive impacts than the suppressive effects caused by policy intensity, both in direct and spillover terms. These findings highlight the importance of inter-regional policy integration and offer new insights for regional environmental governance.

Reducing product fit uncertainty is a critical strategy for retailers to enhance sales and profitability in e-commerce. This paper proposes an optimization-based product sampling strategy to improve sales promotion and profits using recurrent neural networks (RNN) method. Unlike conventional selling modes, the strategy allows consumers to purchase discounted product samples while receiving coupons for subsequent full-priced purchases. A profit optimization model is formulated incorporating constraints such as production costs. The methodological breakthrough lies in introducing RNN to solve this constrained optimization problem, offering a novel approach to dynamic pricing and marketing strategy optimization. In addition, a projection method has been introduced to avoid the additional operation of normalizing product prices in existing methods. The proposed RNN-based framework ensures real-time adaptability, robustness, and efficient convergence, which addresses the complexities of sample distribution and coupon redemption and provides a new perspective on pricing strategies for retailers. The feasibility and effectiveness of the model are demonstrated through numerical simulations, providing valuable insights for retailers seeking promotion-driven pricing strategies.

Multi-class financial distress prediction (FDP) can accurately assess the corporate financial status. Improving its prediction performance is the academic focus. Feature selection and classifier models play a crucial role in the multi-class FDP model. Therefore, this paper proposes a new hybrid feature selection and an improved stacking ensemble model. The hybrid feature selection uses information gain and an improved particle swarm optimization to filter the indicators. The hyperopt hyperparameter optimization method is used to optimize the base learners of stacking ensemble model; The F1-score weighted optimization method is designed for dealing with the discrepancies of the base learners; To objectively solve the combination configuration problem of stacking ensemble model, a constrained genetic algorithm is proposed. The Chinese listed companies are used as research objects for empirical research. The results show that the hybrid feature selection outperforms other feature selection. The F1-score weighted optimized model has 8.97% higher accuracy than the unweighted optimized model. The proposed model performs better in terms of accuracy, robustness, and sensitivity compared to the baseline models and the classifier models in existing multi-class FDP studies. The proposed hybrid feature selection and the improved stacking ensemble model provide new and reliable research ideas for multi-class FDP.

Predicting corporate financial crises is challenging due to the imbalanced and multi-state nature of real-world financial datasets. To address this issue, we developed a reinforced distillation learning (Rf-DL) method that combines fine-grained classification with reinforcement learning and knowledge distillation (KD). Corporate financial conditions can be subdivided into four categories using the Rf-DL method: financial soundness, onset of crisis, moderate crisis, and severe crisis. Deep networks are employed for pre-training, then sample weights are iteratively updated through a reward function before an optimal student network is identified to improve multi-class imbalanced classification. Fine-tuning techniques are also applied to enhance accuracy while reducing the number of parameters. Empirical results based on Chinese listed companies indicate that the Rf-DL model outperforms traditional deep networks and standard KD in recognizing multi-class financial conditions. The reinforced distillation mechanism effectively recognizes heterogeneity within classes, while comparisons with ensemble models highlight its success in handling minority classes. In summary, this paper presents a practical and robust approach to financial crisis prediction.

Financial distress prediction (FDP) is critical for companies, banks, and investors, and artificial neural networks (ANN) have been proven to be an efficient method for FDP. However, the “curse of dimensionality” in FDP not only increases the computational complexity, but also reduces the prediction accuracy. To solve this problem, this paper takes an ANN model as the basic classifier and presents a new two-stage feature selection method integrated with multiple filters and a wrapper method. The financial data of Chinese listed companies are applied for comparative analysis to verify the effectiveness of the constructed method. The results demonstrate that the proposed method achieves a smaller feature subset and better predictive effect than other methods, thus solving the “curse of dimensionality” more effectively and improving the accuracy. In addition, SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDP) are employed to investigate the relative importance of selected features. Their results increase the credibility of the proposed model, giving users more confidence in using this “black box” model.

The motivation of this article is to help financial soundness companies understand their specific financial status so that they can take timely measures to avoid financial distress. Existing multi-class financial distress prediction (FDP) studies have mainly segmented financial crisis status, with less attention paid to financial soundness companies. To fill this gap, we propose a new multi-class definition of FDP from the perspective of financial soundness enterprises. The financial states are defined as financial soundness, moderate financial soundness, mild financial soundness and financial distress. We propose a stacking ensemble model for multi-class FDP. First, deep neural network, multinomial logit regression (MNLogit) and multivariate discriminant analysis models are used as basic classifiers to obtain preliminary prediction results. Second, MNLogit is used to integrate the results from the previous step. To increase the effective information, stock information is then added into the model. The proposed model was trained using data from 2007 to 2019 for Chinese listed companies and tested using data from 2020. The results show that the MacroR-Pre, MacroR-Rec, MacroR-F1 and MacroR-AUC of the proposed model are better compared with the benchmark model, including individuals and ensembles, with 87.05%, 90.68%, 88.70% and 88.20%.The addition of stock information and non-financial indicators can improve the accuracy of the multi-class FDP model by about 8%. The innovativeness of this paper is twofold. First, it proposes a new multi-class definition of enterprise financial status. Second, a multi-class FDP based on stacking is constructed, which provides a new method for solving the multi-class FDP problem. The study shows that the proposed multi-class definition and stacking model are suitable for analysing financial soundness enterprises, which can help managers effectively grasp the specific financial status and have strong practical significance.

One effective solution for companies to expand their market share is to assign customers to optimal marketing strategy, also known as treatments, which are widely employed in randomized controlled trials or A/B tests. However, achieving optimal treatment assignment poses great challenges due to the limitations such as financial constraints and ethical issues associated with randomized controlled trials. To address the challenges, we propose a counterfactual-based uplift modeling approach. This approach involves generating counterfactual treatments and estimating corresponding effects using supervised learning models, ultimately determining the optimal treatment. Our methods have been evaluated on both synthetic and real-world data, demonstrating superior performance compared to other uplift modeling approaches in terms of the Qini coefficient. This study not only contributes to the research on causal inference in the business field but also offers practical implications for companies seeking to enhance business performance through effective marketing treatment assignment.

E-commerce, as a dominant business pattern that integrates online shopping with offline sales, has gained more and more popularity recently. Although this kind of business scheme brings great convenience for daily life, the product fit uncertainty leads to an increase in return rates. This paper first introduces several novel sales models based on different consumer groups including normal purchase model, Try-Before-You-Buy model and group buying under Try-Before-You-Buy model. More specifically, normal purchase model serves as an effective promotional strategy, attracting new customers and generating excitement around products. Try-Before-You-Buy model allows consumers to experience products before purchase and is helpful in finding more potential consumers. Through this model, not only can consumers take advantage of bulk discounts, but also helps retailers anticipate demand more accurately. These business models also introduce challenges such as increased shipping costs, depreciation of returned products, and the emergence of speculators. Then, the above three models are abstracted as related constrained optimization problems. After that, the recurrent neural network approach is applied to solve the constrained optimization problems. Numerical examples are given to show the effectiveness of the proposed neural network.

We investigate the impact of analyst status on recommendation performance. Using 281,886 analyst recommendations from Chinese stock markets from 2007 to 2022, we employ a quasi-experiment approach by leveraging the star analyst election and utilize a time-varying difference-in-differences model to examine the difference in cumulative abnormal returns resulting from analyst recommendations before and after the election. Our findings reveal that when analysts are promoted to star status, the stocks they recommend exhibit improved performance. Interestingly, the star effect is particularly pronounced for analysts working in small brokerage firms and persists even during the economic shocks.

The assessment of credit risk for P2P lending platform applicants is critical to investors. Feature engineering is an essential technique in distilling classification knowledge during the credit risk prediction data preprocessing stage. Although previous literature used feature selection methods to identify key features, feature transformation is more useful in discovering intrinsic nonlinear characteristics in credit data. In this study, we propose a synthetic multiple tree‐based feature transformation method to generate features. Multiple tree‐based feature transformation methods are employed and fused to acquire a new feature set. The bagging‐based tree ensemble feature transformation method (Bagging‐TreeEnsembleFT) and boosting‐based tree ensemble feature transformation method (Boosting‐TreeEnsembleFT) are two types of feature transformation methods that we specifically propose to validate their effect. We verify the credit risk prediction performance using the proposed synthetic feature transformation methods on real P2P Lending credit datasets. Empirical analysis demonstrates that tree‐based ensemble feature transformation methods with boosting ensemble strategy achieve better prediction performance on various datasets corresponding to different partitions and class distributions compared to tree‐based ensemble feature transformation methods with bagging ensemble strategy and individuals. Moreover, the proposed synthetic feature transformation method improves the credit risk prediction performance in terms of accuracy, AUC, and F1‐score.

Accurate forecasting of duty-free shopping demand plays a pivotal role in strategic and operational decision-making processes. Despite the extensive literature on sustainability, operations management, and consumer behavior in the context of duty-free shopping, there is a noticeable absence of an integrated end-to-end solution for precise demand forecasting. Furthermore, existing forecasting models often encounter limitations in effectively leveraging multi-source data as reliable indicators for duty-free shopping demand. To address these gaps, our study introduces a pioneering deep-learning architecture known as the Attention-Aided Interaction-Driven Long Short-Term Memory-Convolutional Neural Network Model (AI-LCM). Designed to capture intricate cross-correlations within multi-source data, encompassing search queries, COVID-19 impact, economic factors, and historical data; this model represents a significant methodological advancement. Rigorous evaluation against state-of-the-art benchmarks conducted on robust real-world datasets confirms the superior forecasting performance exhibited by our AI-LCM model. We elucidate the manifold implications for various stakeholders while illustrating the extensive applicability of our model and its potential to inform data-driven decision-making strategies.

Firms increasingly rely on the third-party platforms to manage customer complaints. In this article, we explore the impact of complaints on firms' idiosyncratic risk by considering both the valence and channel characteristics of complaints. Through textual analysis, complaints are categorized into four types based on the intersection of two dimensions: the valence (mild or severe) of the complaint and the channel (firm or platform) through which the complaint is lodged. Empirical analysis is conducted to test the hypotheses. The results reveal that firms' idiosyncratic risk rises after severe complaints but decreases after mild complaints. Moreover, while mild platform-channel complaints increase risk, severe platform-channel complaints reduce risk. This study highlights the disparate impacts of the channel and valence of complaints on manufacturing and service firms. This research contributes to the literature on online complaints and the marketing-finance interface, offering valuable insights for firms seeking to mitigate their idiosyncratic risk through complaint management strategies.

This paper examines whether auditors of firms with higher financial risk disclose less boilerplate of KAMs. Our empirical study demonstrates that KAMs are more differentiated if the firms are subject to higher financial risk. This association is only pronounced for client firms in the weaker legal environment and audit firms with larger market shares. In addition, client pressure will weaken the motivation of auditors to differentiate KAMs of firms with a higher level of financial risk. We also find that auditors are more likely to differentiate KAMs related to firms’ profitability if firms are subject to high financial risk.