Diverse Algorithmic Recourse In Multi-Criteria Cost Scenario

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/98253

請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/98253

題名:	Diverse Algorithmic Recourse In Multi-Criteria Cost Scenario
作者:	黃尚崴;Hwang, SHANG-WEI
貢獻者:	資訊工程學系
關鍵詞:	演算法
日期:	2025-07-17
上傳時間:	2025-10-17 12:33:06 (UTC+8)
出版者:	國立中央大學
摘要:	當面臨困境時，人們往往會感到迷惘，不確定下一步該如何行動。在這種情況下，許多人會依賴 recourse system（補救系統）來尋找具成本效益的策略，以達成他們的目標。然而，現有的 recourse system 普遍存在一個限制：它們通常僅從單一準則（single-criterion) 的角度出發，缺乏處理問題多維度特性的彈性。因此，若僅針對某一個面向進行優化，可能會忽略其他層面的不利影響，導致原本看似可行的路徑在實際執行時變得困難甚至不切實際。在本研究中，我們提出一種基於圖論 (graph-theoretical) 框架的 recourse system，將問題轉化為一個最短路徑 (shortest path) 問題。為了能同時考量多個準則，我們採用帕雷托最適路徑 (pareto-optimal path) 的方式，找出在整體成本與實際可行性之間達到更佳平衡的解。此外，我們也設計了一種新穎的演算法來提升所產生路徑的多樣性，從而為使用者提供更豐富、具行動性的個人化 recourse 選擇。另外，我們亦結合了計算幾何 (computational geometry) 的概念，開發出一種可降低圖結構複雜度的方法，使得在處理大型資料集時能夠更高效地計算 recourse。 ;When facing challenging situations, such as being denied a bank loan, individuals often struggle to determine the appropriate next steps to improve their circumstances. In such cases, many turn to recourse systems[15, 17, 16] to identify cost-effective strategies for achieving their goals. However, existing recourse systems share a common limitation: they typically operate from a single-criterion perspective, lacking the flexibility to account for multiple dimensions of a problem. As a result, solutions optimized for one criterion may overlook adverse effects from other aspects, leading to unexpectedly difficult or impractical paths to success. In this paper, we propose a recourse system based on a graph-theoretical framework, where the problem is formulated as a shortest path problem. To accommodate multi- ple criteria, we adopt a pareto-optimal path approach[30], enabling the identification of solutions that strike a better balance between overall cost and practical feasibility. Fur- thermore, we introduce a novel algorithm designed to enhance the diversity of the gener- ated paths, thereby offering users a wider range of actionable and personalized recourse options. Additionally, we develop a method that integrates principles from computational ge- ometry to reduce the complexity of the graph, allowing for more efficient recourse com- putation, especially when dealing with large-scale datasets.
顯示於類別:	[資訊工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	8	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....