multi-criteria decision analysis

To address the challenge that conventional supervised decision support systems struggle to reconcile conflicting preferences among multiple stakeholders, this paper proposes a preference-driven participatory decision-making framework. Methodologically, it formalizes decision-making as a contextualized multi-objective optimization problem, introducing (i) a reward-function-based stakeholder representation mechanism, (ii) a model-agnostic and interpretable modular architecture, and (iii) a dynamic trade-off and consensus-generation mechanism integrating compromise functions with user-defined composite scoring. Experiments on two real-world scenarios demonstrate that the framework significantly outperforms pure predictive baselines in balancing competing objectives. Ablation studies confirm its robustness across diverse models, scales, and domains. The core contribution is the first general-purpose, multi-stakeholder decision support paradigm that simultaneously ensures interpretability, flexibility, and consensus orientation.

This paper addresses the challenge of simultaneously aggregating individual preferences and private information in public project decision-making to maximize social welfare while resisting strategic manipulation. We propose a two-stage mechanism: Stage I aggregates Bayesian signals via prediction markets or betting mechanisms; Stage II selects projects using quadratic transfers for preference aggregation. Our work is the first to jointly model both preference and information motives, yielding a fully strategy-proof, incentive-compatible mechanism robust against all forms of manipulation. We derive the first non-asymptotic price-of-anarchy bound for quadratic transfers and prove that the price-of-anarchy converges to 1 in large populations. Under mild assumptions, the mechanism guarantees budget balance and robust price-of-anarchy—i.e., welfare remains close to optimal even under adversarial deviations.

This paper addresses three core challenges in multi-criteria decision-making (MCDM): difficulty in modeling group consensus, neglect of criterion interdependencies, and weak representation of decision-maker preference uncertainty (e.g., normal/triangular distributions, interval-valued preferences). To this end, we propose the first unified Bayesian probabilistic framework for MCDM. Methodologically, we integrate probabilistic graphical models with finite mixture models and introduce a novel hierarchical group identification algorithm to detect homogeneous subgroups, alongside a probabilistic mechanism for ranking criteria and alternatives. Our key contribution lies in establishing a holistic probabilistic modeling paradigm for MCDM, enabling flexible incorporation of diverse preference representations and rigorous handling of interval-valued probabilities. Empirical evaluations demonstrate that our approach significantly improves accuracy in group consensus identification and robustness in ranking outcomes, while outperforming conventional MCDM methods in uncertainty adaptability and statistical interpretability.

TOPSIS suffers from poor interpretability due to its black-box nature—particularly under high-dimensional criteria and heterogeneous criterion weights. Existing multidimensional sensitivity diagram (MSD) visualization tools assume equal criterion weights and lack the capability to model weight sensitivity. This paper pioneers the integration of explainable AI (XAI) principles into classical multi-criteria decision-making (MCDM), proposing a weight–ranking response surface visualization framework. Leveraging gradient-based sensitivity analysis and local linear approximation, combined with systematic parameter-space sampling, the framework enables interactive counterfactual analysis and robustness assessment within a D3.js implementation. Evaluated across multiple benchmark datasets, the method precisely identifies critical weight-perturbation thresholds that trigger rank transitions, thereby substantially enhancing decision-makers’ understanding of and trust in TOPSIS’s internal aggregation logic.

In human-AI collaborative decision-making, tension arises between algorithmic recommendations and human judgment, particularly under data overload, where both observable quantitative objectives and unobservable qualitative preferences (e.g., political feasibility, community acceptability) must be jointly considered. Method: We propose a generative curation framework that dynamically balances optimality and diversity via a dual-path architecture: differentiable sampling and sequential multi-objective optimization. It integrates Gaussian process modeling of latent qualitative utility with a novel diversity metric. Contribution/Results: The framework ensures controllable recommendation set size, near-optimality, and high coverage. Evaluated on policy simulation and operations management tasks, it achieves a 37% improvement in candidate set coverage and a 29% increase in user adoption rate, significantly enhancing decision efficiency and robustness.

This study addresses a key limitation in traditional strategic research, which often treats firms as unitary decision-makers and overlooks the heterogeneity and potential conflicts among individual preferences within organizations. The authors propose a formal mathematical framework that integrates random utility models with specific aggregation mechanisms—such as unanimity or pluralistic governance—to derive, for the first time, an operational organizational utility function. This function is not a simple average of individual utilities but is endogenously shaped by the chosen aggregation rule: unanimity amplifies risk aversion, whereas pluralistic governance fosters risk-seeking behavior. Through applications to Cournot competition and principal–agent models, the framework effectively explains and predicts organizational behavior in strategic interactions, revealing how governance structures fundamentally reshape organizational preferences and influence competitive and cooperative outcomes.

This study addresses the challenge users face in constructing, refining, and validating personalized evaluation criteria when making decisions based on reviews, particularly due to the frequent oversight of infrequent yet critical details. To tackle this issue, the authors propose a visual analytics system that integrates large language models with the Analytic Hierarchy Process (AHP) to automatically generate an initial decision model from user reviews and support human-AI collaborative iterative refinement of criteria, weights, and evidence provenance. The system introduces a novel coverage gap detection mechanism to identify missing evaluation dimensions, incorporates AHP-based consistency constraints for interactive weight adjustment, and enhances decision transparency through multi-level scorecards and exportable reports. Experimental results demonstrate that the system significantly improves users’ control over their evaluation criteria and the overall trustworthiness of their decisions.

This study addresses the challenge posed by heterogeneity in individual values, which often impedes consensus in value-based decision-making. To this end, the paper proposes a decentralized optimization framework that, for the first time, enables the generation of multiple value agreements that accommodate individual differences while reflecting societal diversity. The approach integrates participatory value elicitation with data from the European Values Survey and models individuals’ willingness to compromise to effectively aggregate diverse value systems. Experimental evaluations in two real-world scenarios demonstrate that the proposed method significantly enhances individual utility and outperforms existing aggregation techniques.

This study addresses the problem of simultaneously identifying the piecewise linear additive value functions of two decision makers from anonymous and unlabeled preference responses. By designing a joint preference elicitation mechanism that operates without knowledge of response ownership, the approach uniquely reconstructs both value functions by integrating known breakpoint information, a tailored preference querying strategy, and combinatorial optimization techniques. This work represents the first successful simultaneous identification of multiple decision makers’ additive value functions under an anonymity constraint, thereby overcoming the conventional reliance on explicitly attributed responses in preference modeling. Under noise-free conditions, the method guarantees exact recovery of the complete value functions.

This work addresses the limitation of existing scenario theory, which is confined to single-criterion robustness evaluation and thus ill-suited for real-world decision problems involving multiple criteria that rely on distinct datasets. To overcome this, the paper proposes a general data-driven multi-criteria scenario framework that jointly models violation risks across all criteria, enabling precise quantification of overall robustness guarantees that simultaneously satisfy every criterion. By integrating scenario optimization, probabilistic robustness analysis, and joint risk modeling over multiple datasets, the approach substantially improves the accuracy of robustness certificates. This advancement transcends the constraints of traditional single-criterion methods and offers a theoretically rigorous, scalable, and more accurate framework for delivering joint robustness guarantees in a broad range of multi-criteria decision-making settings.