Abstract

Every day, people and organizations make decisions that involve multiple, often conflicting, criteria. The complexity of these decision-making problems arises from the need to balance competing factors such as fairness, efficiency, and individual preferences. This thesis focuses on two-sided matching (TSM), a particular multi-criteria decision-making (MCDM) problem central to fields like operations research and market design. In TSM systems, two distinct sets of agents—such as employers and job seekers, schools and students, or donors and recipients—seek to be matched with members of the other set. While most research addresses how to match agents, this thesis focuses on the underlying decision criteria by examining how the criteria can be defined in a way that is fair, complete, and free from bias considering all human factors in effect.
Traditionally criteria have been determined by expert opinions, surveys, or historical data. However, these methods may be biased, cannot escape from past inequalities and fail to capture the complete criteria. To address this limitation, this thesis proposes a novel interdisciplinary methodology that integrates Natural Language Processing (NLP), optimization theory, and engineering design science to discover criteria from human-centered problem descriptions. This approach extracts objective decision criteria from natural language descriptions of the matching problem in a systematic and inclusive way using environment-based design (EBD), environment-based life cycle analysis (eLCA), and recursive object model (ROM). The utilization of these engineering design methodologies allows for adaptive, domain-independent, and data-driven criteria formulation, eliminating the need for subjective inputs.
To demonstrate the effectiveness of the proposed method, the methodology is applied in both static and dynamic matching scenarios. Static matching refers to problems involving fixed sets of participants, while dynamic matching accounts for changing sets, such as time-based arrivals. For
iv
the static case, an assignment-based optimization model is used to generate matches based on the identified criteria. For the dynamic case, a perishable capacity management optimization model is adopted, allowing the system to handle uncertainty and last-minute changes effectively.
Key contributions of this research include: a domain-independent, adaptable, and automated methodology for discovering decision criteria in two-sided matching (TSM) problems—eliminating reliance on subjective bias; an original application of design science principles to TSM, framing decision-making as a contextualized, iterative process grounded in stakeholder needs and environmental understanding; a systematic, algorithmic approach for defining input parameters in multi-criteria decision-making (MCDM) optimization models, bridging natural language understanding and mathematical formulation; a significant advancement in fairness and transparency within TSM systems by ensuring that matching decisions are based on objective, inclusive, and human-centered criteria. By redefining how matching problems are formulated and solved, this research lays the foundation for more equitable, adaptive, and human-centered decision-making systems.