Evaluating credit risk is crucial for financial institutions because it determines whether loans should be granted to applicants, impacting potential returns and losses for the institution. However, modeling the heterogeneous information of applicants while simultaneously balancing institutional gains and losses faced by the institution is a complex challenge. To address the issues, this study proposes a credit risk evaluation framework with heterogeneous information: combining three-way decision (TWD) theory and graph sample and aggregate (GraphSAGE) learning. This framework (i) integrates a fuzzy similarity relation based on fuzzy theory, constructing a GraphSAGE model for graph-structured data, (ii) combines the results of GraphSAGE with the TWD theory to divide loan applications into three regions: positive, negative, and boundary, and (iii) adds additional information for instances of the boundary for final evaluations, providing new perspectives for research in credit risk evaluation. The experiment results demonstrate the excellent effectiveness of the method in credit risk evaluation, coping well with data imbalances while balancing losses and gains.

Restaurant selection has become a complex decision-making process for consumers, driven by an overwhelming volume of online reviews. While text and numerical reviews provide valuable insights, the increasing use of visual content, further enriches consumer evaluations. However, existing research lacks effective methods for integrating multimodal reviews to facilitate informed decision-making. To address this gap, this paper proposes a novel approach for restaurant selection based on multimodal online reviews, the contributions of which mainly focus on the following aspects: (i) employ image captioning techniques to convert image review into textual descriptions, bridging the gap between image and text, (ii) apply text analysis methods to extract relevant evaluation criteria from both text and image-generated descriptions, and (iii) integrate insights from both modalities by assessing the object and content consistency between image and text, ensuring the reliability of reviews. The method is applied to Yelp, using a dataset of 31,412 reviews from 10 restaurants. Eight evaluation criteria are extracted from both text and image reviews. The results show that compared with single-modal and dual-modal review-based recommendation methods, the proposed multimodal approach uncovers more comprehensive evaluation criteria and generates more realistic ranking results. Additionally, the proposed information fusion method outperforms traditional fusion methods in effectively integrating multimodal information.

In Farm-to-Supermarket Direct Purchase (FSDP) networks, independent logistics operations cause severe vehicle-cargo mismatches and unsustainable bi-directional costs. Existing collaborative routing research neglects two-stage vehicle-cargo matching and lacks scalable methods to compute multi-party sub-coalition costs simultaneously. To bridge these gaps, we propose an online collaborative pickup and delivery framework. We formulate a bi-directional routing optimization model capable of concurrently generating cost metrics for all sub-coalitions. To efficiently solve this NP-hard problem, we develop an enhanced Adaptive Large Neighborhood Search (ALNS) algorithm integrated with K-means spatial clustering. Subsequently, the Shapley value is applied to guarantee mathematically rigorous cost sharing among participating enterprises. Real-world cases and extensive experiments validate the framework's superiority. The mode reduces total logistics costs by 11.71%-24.60% and driving mileage by 27.33%, yielding an average enterprise cost saving of 25.29%. Furthermore, this approach achieves a critical "subsidy phase-out" effect for governments, maintaining robust economic benefits across diverse rural dispersion scenarios and stringent time-window constraints.

Stage III non-small cell lung cancer (NSCLC) presents complex challenges in treatment decisions due to extensive disease characteristics and patient heterogeneity. Multidisciplinary teams (MDT) offer personalized treatments by integrating diverse expertise. However, resource and time constraints limit MDT practicality in healthcare. Addressing this, we propose an interpretable approach for intelligent MDT treatment recommendations. Our approach focuses on enhancing clinical text representation and the interpretability of recommendations. Using a dual-level embedding technique, the local and global textual information can be captured. By delving into intricate factors at word, phrase, and sentence levels, we explain treatment recommendations for heterogeneous patients. Specially, attention flow is employed to consider the relationship between attentions across multiple layers, and attention mechanisms screen out important words and sentences. Our method achieves over 85% across accuracy, precision, recall, and F1 score in stage III NSCLC treatment recommendations. Furthermore, we assess the association between our proposed model and patient survival outcomes by analyzing patients who did not undergo MDT consultation. The results demonstrate that patients receiving model-concordant treatments exhibited significantly higher survival rates at 1, 3, and 5 years compared to those receiving model-nonconcordant treatments. Moreover, Kaplan-Meier survival curves confirm the improvement in survival outcomes associated with model-concordant treatments. Additionally, ablation analysis validates the rationality of the model structure.

Dynamic organizational networks are characterized by evolving collaborations, hierarchical structures, and cross-departmental goal conflicts, which challenge existing community discovery and resource allocation methods. To address these issues, we propose MCHH-TF, a dynamic optimization framework that integrates adaptive time-window segmentation (ATWS), curvature-adaptive hyperbolic space embedding (HSE), a fourth-order man agement coupling tensor (M-Tensor) for hierarchical-temporal conflict modeling, and a stability-constrained NSGA-II procedure to jointly optimize community quality, cross-layer conflict, and allocation cost. Experimental validation on six real-world and large-scale network datasets demonstrates that MCHH-TF consistently outper forms state-of-the-art approaches. It achieves high-quality community partitioning, with an average modularity of about Qh approximate to 0.79 and an improvement of approximately 8-10% over the strongest competitive baselines. Meanwhile, MCHH-TF reduces cross-layer conflict intensity by about 25-35% and lowers resource allocation cost by about 35-45%, while maintaining lower runtime and stable F1-score performance across datasets. Moreover, MCHH-TF remains robust under noise interference, showing only limited performance degrada tion under strong perturbations. These results indicate that MCHH-TF offers an effective decision-support tool for enterprise resource allocation and cross-department collaboration optimization in dynamic organizational settings.

Online restaurant reviews often express multiple viewpoints within a single review, yet existing studies predominantly focus on aggregated representations at the restaurant level and rarely model the relative importance of opinions at the single-review level. To bridge this gap, we propose a novel opinion importance analysis model that jointly considers content semantics and sentiment characteristics to estimate the relative importance of viewpoints expressed within a single review. Building on this, we develop a single-review level opinion representation framework, together with an opinion completion strategy that infers missing viewpoints based on neighboring reviews, and an importance-aware aggregation mechanism for integrating information across reviews. Experiments on a real-world Yelp dataset comprising 31,412 reviews from 10 restaurants extract six key themes using a combined topic model. Empirical analysis shows that multi-theme opinion expression within individual reviews is prevalent, with approximately 56%-86% of reviews containing viewpoints across multiple themes, while a substantial portion of theme-level opinions remains unobserved within individual reviews. The proposed opinion completion strategy effectively recovers these missing viewpoints, enabling more comprehensive review representations. Comparative evaluations demonstrate that explicitly modeling opinion importance and completing missing viewpoints yield more informative review representations than existing alternatives. Sensitivity analysis further confirms the robustness of the proposed framework with respect to key parameters. This work provides a novel paradigm for review analysis, shifting from coarse-grained aggregation at the restaurant level to fine-grained, singlereview opinion modeling that captures both opinion importance and missing viewpoints, with important implications for advanced opinion mining and personalized recommendation systems.

Grey plastic is a representative traditional architectural decoration craft in the Lingnan region in China, carrying rich historical and cultural values as well as distinctive regional artistic characteristics. However, the grey plastic craft is currently facing problems such as inheritance gaps and a shortage of craftsmen, and its restoration projects impose extremely high professional requirements on contractors. Existing contractor selection methods are mostly applicable to ordinary construction projects and are difficult to adapt to its particularity, which may easily lead to risks such as substandard restoration quality. Therefore, this paper proposes a contractor selection method for grey plastic decoration projects of cultural relic buildings based on the BWM-TODIM method. Firstly, an evaluation system covering six core criteria is constructed; secondly, the BWM is adopted to determine the criteria weights; thirdly, the TODIM method is used to characterize the decision-makers' loss aversion psychology and rank the candidate contractors; finally, an empirical analysis is conducted with a grey plastic restoration project in Lingnan as a case to verify the feasibility and effectiveness of the method. This study can provide decision support for the scientific selection of contractors for grey plastic decoration projects and contribute to the sustainable protection of cultural heritage. The scope of this study is limited to contractor selection for grey plaster decoration engineering of cultural relic buildings.

Healthcare insurance fraud poses significant threats to global medical security systems, causing substantial financial losses and eroding public trust.Existing detection methods often fail to capture complex, multiple behavioral fraud patterns due to their reliance on single-relation networks or staticfeature engineering. To address these limitations, we propose the Multi-ViewReinforcement Contrastive Learning (MV-RCL) framework, which integratesthree tightly coupled components: (1) Four fraud-specific relational graphviews, corresponding to Fragmented Hospitalization, Overutilization, Overdiagnosis, and Upcoding behaviors, that disentangle heterogeneous behavioral patterns into behavior-discriminative provider–provider representations;(2) A Bernoulli Multi-Armed Bandit-driven dynamic neighbor sampling module that adaptively determines per-node sampling quantities to suppress aggregation noise from irrelevant connections; and (3) A multi-view contrastivelearning mechanism enforcing intra-view, inter-view, and global representation consistency to improve embedding separability between fraudulent andlegitimate providers. Evaluated on a Medicare claims dataset comprising541,071 claims, MV-RCL outperforms 10 state-of-the-art baselines spanningclassical classifiers, homogeneous and heterogeneous GNNs. Interpretabilityanalysis provides evidence-grounded attribution for both isolated and collaborative fraud patterns. MV-RCL thus offers an accurate, stable, and interpretable detection tool capable of modeling complex multi-behavioral fraudpatterns, advancing graph-based methods for real-world healthcare insuranceoversight.

The limitations of traditional multi-attribute decision-making (MADM) have become increasingly evident, highlighting the need for intelligent MADM. The Elimination and Choice Translating Reality (ELECTRE) III MADM method suffers from tedious calculation caused by pairwise comparison, and cannot effectively determine the indifference, preference, and veto thresholds. Recently, neural networks (NNs) have been widely used to address challenging MADM problems and develop intelligent methods. This study, therefore, focuses on building an intelligent ELECTRE III MADM model based on NNs for intelligent computation, assignment, and decisionmaking. In this case, a multiprocessing calculation algorithm is designed to enable ELECTRE III to perform efficiently in high-volume data environments. Further, an NN-based detection algorithm is designed to detect the threshold parameters of ELECTRE III. Finally, the application of the established intelligent ELECTRE III model is demonstrated using a real-world dataset from QS World University Rankings. The model is shown to have good stability, strong computing power, flexible applicability, and the ability to support practical decision-making.

Medical Entity Normalization (MEN) aims to map informal clinical mentions to formal medical concepts in certain terminologies. Despite the rapid development of MEN research in English, Chinese MEN presents unique challenges, including (1) semantic confusion arising from missing word delimiters and compound terms, (2) limited annotated data with context-deficient texts, and (3) the multi-implication issue where a single mention may correspond to multiple formal concepts. In this study, we propose a novel Chinese MEN framework that integrates multi-granularity semantic extraction and multi-task fusion to tackle these challenges. First, we introduce a word-lattice structure to capture rich word-level semantics of Chinese entities while mitigating segmentation errors. Second, we employ a pre-trained medical language model to encode character-level semantics with reduced data reliance, enhanced by adversarial training for robust few-shot fine-tuning. Third, we utilize a multi-task fusion framework to jointly model implication number prediction and mention-concept matching, effectively addressing the multi-implication issue. By incorporating multi-similarity loss to guide online hard negative mining, the multi-task training process effectively captures discriminative features of ambiguous concepts. Our model is evaluated on clinical entity normalization datasets from the CHIP 2019 and 2020, achieving high accuracy of 91.54% for procedures and 68.2% for diagnoses, significantly surpassing other baseline methods. The suggested model is expected to disambiguate Chinese medical terms, providing a solid foundation for downstream clinical applications.

Community detection in organizational networks is vital for optimizing team structures, yet existing methods face critical challenges: Static models ignore temporal dynamics, dynamic single-layer approaches overlook cross-layer interactions, and multi-objective frameworks often optimize goals in isolation, leading to suboptimal real-world performance. We propose the Multi-Objective Dynamic Multi-Layer Hypergraph Modeling Framework (MO-DMLHM), integrating three innovations: (1) Adaptive Dynamic Hypergraph Modeling with dual-scale decay and adaptive time windowing to capture spatiotemporal dynamics; (2) Four-Dimensional Multi-Objective Optimization balancing modularity, cross-layer consistency, stability, and efficiency via Pareto-optimal NSGA-III; (3) Hybrid Encoding Evolutionary Algorithm jointly optimizing hyperedge activation and node membership through spectral clustering-guided mutation and betweenness centrality-driven crossover. Experiments on diverse organizational networks show MO-DMLHM outperforms state-of-the-art methods in detection accuracy, cross-layer alignment, and stability, reducing coordination costs by nearly 40%. Ablation studies confirm the necessity of dynamic modeling, multi-objective optimization, and hybrid encoding. MO-DMLHM resolves structural-community decoupling in dynamic multi-layer systems, advancing complex network analysis and enabling adaptive governance in organizations, with extensions to smart cities, biological networks, and financial risk management.

In engineering R&D organizations, resource contention among autonomous projects creates complex scheduling challenges in distributed multi-project environments. Traditional coordination mechanisms often prioritize global efficiency while overlooking inter-project fairness, which can lead to stakeholder resistance and execution delays. This study proposes a transparent decision support approach for the Distributed Resource-Constrained Multi-Project Scheduling Problem (DRCMPSP). We develop a two-stage scheduling mechanism that integrates Multi-Criteria Decision-Making (MCDM) into the global coordination process. First, an enhanced Chaotic Genetic Algorithm (CGA) with elitism generates local schedules. Second, global resource conflicts are resolved using MCDM methods. Inter-project fairness is implemented by limiting each project’s relative objective deterioration and by evaluating the dispersion of the resulting project-level burdens. Validation through an Unmanned Aerial Vehicle R&D case study and extensive experiments shows that the TOPSIS-based mechanism achieves a significantly lower standard deviation than the auction-based mechanism under high resource contention. The approach supports transparent and fairness-aware conflict resolution by making trade-offs among project-level outcomes explicit to managers.

The evaluation of circular economy (CE) in iron and steel industry serves to offer recommendations for the sustainable development of this sector. However, the existing evaluation exhibit limitations, including the neglect of information reliability and the subjectivity inherent in modernization processes. Therefore, this paper proposes a comprehensive Z-Entropy-Text Mining-MARCOS method (ZETMM) for evaluating CE in the iron and steel industry. Firstly, to resolve the issue of overlooked information reliability in previous studies, Z-numbers are employed to characterize CE-related data, and an improved comprehensive weighted distance measure for Z-numbers is proposed. Secondly, the innovative framework integrates a novel Z-number weighting method and Text Mining techniques to combine quantitative and qualitative data, thereby reflecting CE priorities as perceived by media outlets and governmental authorities. Additionally, the Z-MARCOS method is further proposed to effectively address the prevalent uncertainty in CE data. Demonstrated in a case study involving seven Chinese provinces, the proposed method demonstrates its practicality and robustness in real-world applications. The results show that Zhejiang, Shanghai and Guangdong achieved strong performance in the iron and steel industry’s CE development in 2019. Based on the results, targeted recommendations are put forward to promote CE advancement. This paper proposes an improved Z-number comprehensive weighted distance measure, a novel index weight calculation method and a new evaluation method under the environment of information with limited reliability. These contributions aim to provide scientific and reliable results for CE evaluation in the iron and steel industry.

Modelling patient trajectories from longitudinal electronic health records (EHRs) is crucial for early chronic disease prediction. Foundation models (FMs), benefiting from the computational power and generalization abilities, offer a promising direction towards understanding patient health progression. However, key challenges of adopting FMs in clinical decisions remain in (1) incorporating multi-modal EHR data into an FM effectively for unified patient representations and (2) ensuring model generalizability across various clinical domains with distribution shifts. To address these challenges, we propose MsHeCare, an FM-based, two-stage learning framework integrating multi-modal representation learning and multi-source domain adaptation (MSDA). In the pretraining stage, MsHeCare performs self-supervised contrastive learning and masked language modelling tasks to mitigate semantic biases across text-based diagnostic and treatment sequences. A cross-attention mechanism further fuses these temporal features with structured static demographic information, enhancing personalized patient representations. In the fine-tuning stage, MsHeCare incorporates an MSDA framework with a novel source importance estimation method, facilitating adaptive knowledge transfer across domains and improving model generalizability. Experiments on two real-world EHR datasets demonstrate that MsHeCare significantly outperforms single-domain baselines and state-of-the-art MSDA methods. Furthermore, we validate the robustness of MsHeCare across varying targetdomain data sizes and demonstrate its alignment with clinical practices through case studies, underscoring its potential for multi-modal, domain-adaptive predictive healthcare systems.

Assessing energy transition paths is critical for sustainable development under global climate change. However, existing studies generally lack an integrated approach combining large-scale public data with multi-criteria decision-making analysis, constraining insight into societal preferences for energy strategies. To address this gap, this study proposes a data-driven multi-criteria decision support framework based on social media text to systematically mine and evaluate energy transition paths. The framework aligns user comments with defined transition paths through a path definition approach including path-comment alignment and discourse-based referential inference, successfully mapping about seventy percent of collected comments and addressing ambiguities in path mentions. An unsupervised criterion extraction model, combining dependency parsing with orthogonality constraints, identifies six coherent and distinct latent evaluation criteria. Probabilistic clustering further reveals four distinct attitudinal groups with differentiated transition strategy preferences. By applying probabilistic linguistic term sets, the framework quantifies the perspectives of each group across all criteria for each path and derives corresponding rankings of transition paths. This study contributes to enhancing public participation in energy decision-making, offers quantitative support for multi-criteria models using social data, and advances the use of social media in energy policy making.

Compared with traditional fuzzy numbers, using Z-numbers to illustrate fuzzy events offers two key advantages. It makes fuzzy events more intuitive for decision-makers, and the second component of Z-numbers acts as a measure of the reliability of the first component. While significant progress has been made on Z-numbers, from the theoretical and practical perspectives, some gaps remain. For example, scholars have seldom focused on the likelihood of Z-numbers, most existing decision-making methods with Z-numbers rarely consider the consensus-reaching processes, and little has been reported on the superior ordering methods in the Z-number environment. To overcome these limitations, first, the likelihood of Z-numbers is defined in combination with the preference ranking organization method for enrichment evaluations (PROMETHEE) type V preference function. Second, the ordering rules for PROMETHEE are discussed. Then, a procedure of the feedback-adjustment method is introduced to help the consensus level of group-alternative ranking reach the threshold. On these bases, an extended PROMETHEE multi-criteria group decision-making method with Z-numbers is proposed. Finally, to verify the feasibility and effectiveness of the proposed method, we examined an intelligent medical-diagnostic-system selection problem and conducted a comparison analysis. We applied the Z-number PROMETHEE approach to a challenging case study requiring a dual-data-driven application. Furthermore, the study suggests future directions for improving the proposed framework in other related contexts.

Medical insurance fraud detection is crucial for minimizing the depletion of insurance pools. While medical expense records (MER) are valuable for this task, their limited availability is often overlooked. Extant studies directly input MER into high-dimensional machine learning (ML) models to achieve fraud detection. Another class of models generates fraud prediction based on probability modelling of samples. To explore whether these two different classes of models can cross-fertilizer each other, this paper incorporates Bayesian network (BN) into extreme gradient boosting (XGB). After obtaining the healthcare fraud predictions, this study employs these results to risk management decisions for minimizing the cost of Medical Insurance Bureaus. To make the optimal cost-based decision, we develop an instance-dependent cost-sensitive XGB (ICXGB) method. Using real-world data, we construct various variables based on the famous Recency, Frequency and Monetary (RFM) principle and empirically assess the prediction performance of probabilistic model and ML. The integrative model shows a significant improvement over BN and ICXGB. Finally, a post-hoc explanation method is adopted to quantify the contributions of the predictors and obtain some management implications.

The dramatic increase in carbon dioxide emissions is a major cause of global warming and climate change, posing a serious threat to human development and profoundly affecting the global ecosystem. Currently, carbon dioxide emissions prediction studies rely heavily on a large amount of data support, and the accuracy of predictions is greatly reduced when data are scarce. In addition, the inherent uncertainty, volatility, and complexity of CO2 emission data further exacerbate the challenge of accurate prediction. To address these issues, a novel hybrid model for CO2 emission prediction is proposed in this paper. A feature screening method is designed for effective and reliable feature selection from the perspective of algorithm stability, which can improve the prediction performance. In order to accurately predict periodic sequences with limited training samples, a least squares support vector machine is employed in this paper. In addition, the parameters of the prediction model are optimised using the improved sparrow search algorithm and enhanced by Sin chaos mapping, adaptive inertia weights and Cauchy-Gauss variables. An empirical study is conducted using Chinese carbon emission data as a case study, and the validity and superiority of the proposed model are verified through comparative experiments. The results show that the improved SSA has stronger global optimisation capability and faster convergence speed. In addition, in terms of prediction results, the hybrid model has the best consistency with the actual data, which significantly improves the prediction accuracy.

In user-generated content (UGC) on online automotive platforms, consumers typically express their evaluations in the form of numerical ratings and textual reviews. To gain deeper insights into consumer preferences regarding new energy vehicles, it is essential to extract personalized individual semantics from UGC. Here, this study proposed a novel method motivated by UGC characteristics for personalized individual semantic (PIS) analysis. The developed method considers both group heterogeneity and individual consistency between the numerical and linguistic evaluations of vehicles in terms of different criteria. Based on the linguistic distribution assessments converted from UGC, we used k-nearest neighbor clustering to aggregate group opinions. Then, two optimization models were constructed based on maximum group heterogeneity and minimum information deviation to model the PISs of these groups. A comprehensive optimization model was also established to assign PISs to flexibly manage various scenarios. To demonstrate the effectiveness and applicability of the proposed model, this study conducted a case study and comparative analysis with evidence from the Pacific Automotive website ( www.pcauto.com.cn ). The results indicated that the proposed method can effectively reveal the personalized semantic preferences of individual and group buyers, with reference value for potential consumers and enterprises.

Credit rating serves as a crucial instrument for lenders to evaluate borrowers' creditworthiness and mitigate the risk of nonperforming loans. However, credit rating tasks often face significant challenges due to multiclass distributions and severe class imbalances. Given the advantages of ensemble learning methods in addressing these challenges, this study presents a novel multiclass imbalance classification framework that integrates the Error Correcting Output Codes (ECOC) decomposition approach with diverse dichotomizer imbalance algorithms to enhance credit ratings. Nevertheless, selecting and quantifying the uncertainty of dichotomizer sets poses challenges. To this end, we introduce a dynamic ensemble selection strategy and evidence theory within the ECOC setup. By tailoring specific dichotomizers to individual samples and consolidating uncertain binary outcomes using belief functions, a resilient ensemble classifier is developed. Extensive experiments on nine KEEL benchmark datasets and two real credit datasets demonstrate its effectiveness in handling severe imbalance in credit rating tasks.