Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90005
Title: | 各種多臂賭博機演算法在個人化推薦之效率、效能與穩健性比較 A Comparison of Efficiency, Effectiveness, and Robustness of Multi-Armed Bandit Algorithms for Personalized Recommendations |
Authors: | 蔡立忠 Li-Chung Tsai |
Advisor: | 黃從仁 Tsung-Ren Huang |
Keyword: | 推薦系統,強化學習,多臂賭博機, Recommendation systems,Reinforcement Learning,Multi-armed bandit, |
Publication Year : | 2023 |
Degree: | 碩士 |
Abstract: | 推薦系統(recommendation system)在當今數位時代扮演著重要的角色。它們是商業領域中的關鍵技術,被廣泛應用於電子商務、社交媒體、音樂和影視娛樂等領域,它們能夠提供個性化的體驗,提高銷售轉換率,增加用戶參與度,促進跨銷和交叉銷售,並增加競爭力。隨著數據和機器學習(machine learning)技術的不斷進步,推薦系統在未來將繼續發揮重要的作用,並不斷優化和創新,以滿足用戶和商業需求的不斷變化。
因此,演算法的評估方式是否能夠有效模擬現實情境,顯得格外重要。過去的演算法評估方式大多基於線下資料集,然而這些方法與現實世界中的用戶互動存在一定差距。因此,為了更貼近真實環境,本研究除了使用傳統非機率性預測方式,同時採用了多臂賭博機(Multi-Armed Bandit)演算法並加入機率性模擬的預測方式進行兩者比較。 本研究同時針對不同的商業目的,將演算法的目標分為前期效率(efficiency)、長期效能(effectiveness)、跨情境穩健性(robustness)三大指標進行比較。通過這種機率性模擬的方式,我們能夠更好地模擬用戶的真實反應。這種模擬方式能夠重複推薦相同的產品,並且更全面地評估演算法的效果。與過去的評估方式相比,我們的研究結果顯示出截然不同且更真實現實的結果。最終本研究將給予各種資料庫特性與商業目的下,建議的演算法,以提供各大平台參考,利於商業發展。 Recommendation systems are vital in the digital era, powering e-commerce, social media, music streaming, and more. They enhance user experiences, drive conversions, and promote cross-selling. Evaluating these systems is crucial, and our study compares algorithms using probabilistic simulation for real-world conditions. We assess efficiency, effectiveness, and robustness, essential metrics for diverse business objectives. By employing probabilistic simulation, we try to simulate user responses in real life and evaluate algorithm performance comprehensively. This approach enables repeated recommendations and yields distinct, realistic results compared to traditional evaluation methods. Our research provides valuable insights for algorithm selection, considering database characteristics and specific business goals. These findings empower major platforms to optimize their recommendation systems and drive business growth. In summary, our study demonstrates the importance of accurately evaluating recommendation algorithms in real-world contexts, highlighting the benefits of probabilistic simulation for improving system performance and user satisfaction. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/90005 |
DOI: | 10.6342/NTU202301110 |
Fulltext Rights: | 未授權 |
Appears in Collections: | 統計碩士學位學程 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-111-2.pdf Restricted Access | 2.25 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.