利用效用函數刻劃使用者偏好—以C2C 平台推薦系統為例

Abstract

推薦系統在電子商務中扮演著重要角色，幫助用戶快速找到所需商品或服務。隨著大數據和機器學習的進步，推薦系統的準確率和效率已有顯著提升。然而，現有推薦系統仍然存在個人化程度不足和推薦清單單調等問題。 為解決這些問題，本研究引入效用函數以刻劃用戶偏好。透過效用函數，我們結合用戶行為和商品特徵，提供更個性化的推薦，同時協助用戶探索可能感興趣的商品。 研究中，我們分析電商平台的瀏覽資料，利用馬可夫鏈蒙地卡羅(MCMC) 演算法估計效用函數，從而獲得每個用戶的偏好參數。與過去的研究相比，我們提出了對參數的解讀方法，深入了解每個用戶的特性，在後續進行其他分析中能提供不小的幫助。 此外，我們提出了三種版面配置方法，並分析了各自的優缺點，為平台方提供了針對不同用戶制定推薦策略的指引。透過本研究，我們期望能提升推薦系統的個人化程度並改善用戶體驗，同時為電子商務平台提供有益的參考，進一步提高推薦系統的準確性和效能。

Recommendation systems play a crucial role in e-commerce, facilitating users in swiftly discovering desired products or services. With advancements in big data and machine learning, recommendation systems have witnessed significant improvements in accuracy and efficiency. However, existing recommendation systems still suffer from issues such as insufficient personalization and monotonous recommendation lists. To address these challenges, this study introduces utility functions to characterize user preferences. By leveraging utility functions, we amalgamate user behavior and product features to provide more personalized recommendations, while assisting users in exploring potentially interesting items. In this research, we analyze browsing data from e-commerce platforms and employ the Markov Chain Monte Carlo (MCMC) algorithm to estimate utility functions, thereby obtaining preference parameters for each user. Unlike prior studies, we propose an interpretation method for these parameters, gaining deeper insights into individual user characteristics and offering valuable guidance for further analyses. Furthermore, we propose three layout configuration methods and analyze their respective pros and cons, offering platform administrators guidance to formulate tailored recommendation strategies for different user segments. Through this study, we aim to enhance the personalization of recommendation systems and improve user experience, while providing beneficial insights for e-commerce platforms to further enhance the accuracy and efficiency of their recommendation systems.