| 基于影子价格的平台推荐系统冷启动算法 |
| A Shadow Price-Based Cold Start Algorithm for Platform Recommendation System |
| 摘要点击 6 全文点击 0 投稿时间：2024-10-10 修订日期：2025-07-19 |
| 查看/发表评论 下载PDF阅读器 |
| 中文关键词 推荐系统；冷启动；平台运营；学习算法 |
| 英文关键词 recommendation system; cold start; platform operation; learning algorithm |
| 基金项目 国家自然科学基金项目（面上项目，重点项目，重大项目） |
| 投稿方向 |
| 作者 | 单位 | 邮编 | | 金凯瑞 | 复旦大学管理学院 | 200433 | | 田林* | 复旦大学管理学院 | 200433 | | 徐以汎 | 复旦大学管理学院 | | | 唐百川 | 亚马逊云科技 | |
|
| 中文摘要 |
| 众多在线平台不断涌现, 商家数量增长和消费者需求多样化使得推荐系统变得至关重要. 新商家冷启动问题, 即由于缺乏历史数据, 使得平台难以预测消费者对新商家的偏好, 进而影响对其的有效推荐, 造成新商家流失, 并损害平台长期收益. 针对推荐多个商家的场景, 建立了冷启动优化模型, 并结合转化率预测、影子价格求以及多臂老虎机模型, 构建了一种基于影子价格的冷启动算法. 结果表明, 该算法可显著提高冷启动成功率以及平台的整体收益. 同时, 该算法使用影子价格的概念解释了算法原理, 为机器学习与运筹优化的结合提供了新的视角. |
| 英文摘要 |
| The proliferation of online platforms, coupled with the increasing number of merchants and the diversification of consumer demands, render recommendation systems crucial. The cold start problem for new merchants, which arises due to the lack of historical data, makes it difficult for platforms to predict consumer preferences for new merchants. This, in turn, affects the effectiveness of recommendations, leading to merchant loss and long-term revenue loss for the platform. To address the scenario of recommending multiple merchants, a cold start optimization model was established. This model integrates conversion rate prediction, shadow price computation, and the multi-armed bandit model to develop a shadow price-based cold start algorithm. Results demonstrate that this algorithm significantly enhances the success rate of cold starts and overall platform revenue. Additionally, the algorithm employs the concept of shadow price to elucidate its principles, offering a novel perspective on the integration of machine learning and operations research optimization. |
| 关闭 |
|
|
|
|
|
