To enhance the accuracy and reliability of large language models (LLMs) in regulatory question-answering tasks, this study addresses the complexity and domain-specificity of regulatory texts by designing a retrieval-augmented generation (RAG) testing framework. It proposes a dimensionality reduction-based semantic similarity measurement method and a retrieval optimization approach leveraging information reasoning. Through the construction of the technical route of the intelligent knowledge management system, the semantic understanding capabilities of multiple mainstream embedding models in the text matching of financial regulations are systematically evaluated. The workflow encompasses data processing, knowledge base construction, embedding model selection, vectorization, recall parameter analysis, and retrieval performance benchmarking. Furthermore, the study innovatively introduces a multidimensional scaling (MDS) based semantic similarity measurement method and a question-reasoning processing technique. Compared to traditional cosine similarity (CS) metrics, these methods significantly improved recall accuracy. Experimental results demonstrate that, under the RAG testing framework, the mxbai-embed-large embedding model combined with MDS similarity calculation, Top-k recall, and information reasoning effectively addresses core challenges such as the structuring of regulatory texts and the generalization of domain-specific terminology. This approach provides a reusable technical solution for optimizing semantic matching in vertical-domain RAG systems, particularly for MDSs such as law and finance.

As an enhanced version of traditional linguistic term sets, Probabilistic Linguistic Term Sets (PLTS) incorporate probabilistic information, thereby offering a more robust approach to Multiple Attribute Group Decision-Making (MAGDM) and significantly improving its efficacy. This paper proposes two novel information aggregation operators for PLTS to address MAGDM problems in the PLTS context. Firstly, we introduce Choquet integral-based generalized arithmetic and geometric operators, which are designed to fuse decision information expressed by different PLTSs, thereby more comprehensively considering the interrelationships among various attributes. Subsequently, we further define measures of group consistency and inconsistency for individual decision information in MAGDM, which are used to determine the information weights of decision-makers. Finally, the group decision information is aggregated using the proposed PLTS aggregation operators. The effectiveness as well as the applicability of the developed method are illustrated through numerical examples and comparative analysis.

Whether the New Energy Vehicle Promotion Policy (NEVPP) enhances supply chain resilience is pivotal to China’s green transition and global industrial security. Using data on A-share listed automobile manufacturers from 2012 to 2024, this study employs a multi-period difference-in-differences approach to identify the policy’s impact. Results show that NEVPP significantly strengthens supply chain resilience, and the findings remain robust across alternative specifications. Mechanism analysis reveals that the policy raises managerial attention, eases financing constraints, and stimulates technological innovation, thereby enhancing resilience through managerial, financial, and technological channels. Heterogeneity analysis by ownership, geography, R&D intensity, analyst coverage, and institutional ownership shows that the effect is stronger for state-owned enterprises, firms in central and western regions, low-R&D firms, those without analyst coverage, those with high analyst attention, and firms with low institutional ownership. This study provides firm-level evidence on the economic consequences of NEVPP, advances understanding of industrial policy and corporate resilience, and offers policy implications for supporting the global energy transition and safeguarding supply chain stability.

Post-disaster emergency logistics is often organized according to prespecified response periods, although operational relief tasks may be completed earlier or later than those calendar boundaries. This study develops a task-completion-triggered multi-stage resource-allocation mechanism (TCMS) for a two-echelon emergency logistics network. Effective first-contact coverage and cumulative forward deployment generate three sequential operational stages—rapid coverage, intensive deployment, and recovery support—through a one-period state-update rule. The operational stage determines the priority regime used to select among feasible allocation schedules. The mechanism is evaluated within a common direct-decision rolling-horizon mixed-integer linear programming (MILP) framework. At each decision epoch, policy-neutral cost-service candidates are generated using total cost and cumulative raw weighted backlog, whereas the current operational stage supplies stage-dependent priority coefficients for selecting the implemented action. TCMS and a fixed-stage baseline therefore differ in the source of their operational stage, while a stage-free rolling-horizon benchmark retains a neutral priority regime. Computational experiments include formulation validation, Base and Medium rolling-horizon policy comparisons, a Large static MILP reference, and trigger-parameter and stage-priority-coefficient sensitivity analyses. The results show that endogenous stage timing changes the selected cost-service operating point, with effects varying across instances. Sensitivity analysis further shows that transition timing and allocation outcomes can vary with trigger settings and stage-priority calibration. TCMS is particularly suited to settings in which transition timing cannot be specified reliably in advance, and managers require an observable basis for adjusting allocation priorities as relief operations progress.

Digital identity and battery-passport systems can make electric-vehicle battery lifecycle records traceable, but traceability alone does not guarantee verifiable evidence or physical recycling compliance. We develop a normative two-stage Stackelberg model in which a regulator chooses audit intensity and verified incentives and a representative obligated firm jointly chooses pre-verification evidence quality and recycling compliance. The benchmark is stress-tested under reduced eligibility-confirmation effectiveness, alternative evidence–recycling interaction, decentralized manufacturer–recycler decisions, and implementation costs. Baseline traceability can leave evidence quality at the reporting floor and compliance incomplete, while additional responsibility exposure can conditionally increase recycling compliance while reducing evidence quality. Coordinated verified incentives improve both decisions and welfare in the benchmark complementarity domain. At zero eligibility-confirmation effectiveness, the fixed positive-rate package can become marginally welfare-inferior under finite capacity; adverse interaction can reverse a cross-effect, decentralization creates coordination underinvestment, and high payment or startup costs create single- or no-positive-payment regions. The model is therefore a forward-looking governance analysis rather than an empirical evaluation or legal calibration of current Chinese or European Union rules.