کنفرانس بین المللی یادگیری و یاددهی الکترونیکی ایران

صفحه اصلی / نوزدهمین کنفرانس ملی و سیزدهمین کنفرانس بین المللی یادگیری و یاددهی الکترونیکی ایران

The Algorithmic Price of Fairness: Moving Beyond First-Come-First-Served in Educational Matchings

نویسندگان :

Maedeh Mosharraf¹

1- دانشگاه شهید بهشتی

کلمات کلیدی :

Exponential Cost-Minimization،Genetic Algorithm،Algorithmic Fairness،Educational Resource Allocation،Stable Matching،Student Satisfaction

چکیده :

The First-Come-First-Served (FCFS) paradigm, while simple to implement, often leads to perceptions of unfairness and suboptimal outcomes in educational resource allocation. This study investigates the algorithmic price of fairness—the trade-off between overall efficiency and equitable outcomes—in the context of assigning research topics to student groups. We model this challenge as a bipartite matching problem and propose two computational solutions: an exponential cost-minimization model, which imposes non-linear penalties to discourage low-priority assignments and promote distributive justice, and a genetic algorithm metaheuristic for global optimization. Through a field experiment conducted over two semesters in a research methods course, we empirically evaluated these models against a baseline FCFS approach. Our results demonstrate a quantifiable price for fairness: while both algorithmic approaches significantly increased overall satisfaction and perceived fairness compared to FCFS, the exponential cost-minimization method achieved a superior balance, yielding higher minimum satisfaction levels with a negligible reduction in aggregate utility. Furthermore, we found that algorithmic transparency—disclosing both peer preferences and the allocation logic—was a critical factor in enhancing perceived fairness, often mitigating the subjective cost of receiving a lower-preference topic. This work provides a framework for intentionally designing allocation mechanisms that navigate the efficiency-fairness trade-off, moving educational practices beyond manual or arbitrary methods toward principled, computational solutions.