| 长期背景下基于多智能体建模的多利益相关者推荐策略研究 |
| The Longitudinal Research on Multistakeholder Recommendation Strategies based on Agent-based Modelling |
| 摘要点击 6 全文点击 0 投稿时间：2024-06-04 修订日期：2025-07-21 |
| 查看/发表评论 下载PDF阅读器 |
| 中文关键词 推荐策略；多智能体建模；公平性；多利益相关者；动态表现 |
| 英文关键词 recommendation strategies；agent-based modelling；fairness；multistakeholder；dynamic performance |
| 基金项目 国家自然科学基金项目（72171165）；教育部人文社科研究资助项目（21YJA630021） |
| 投稿方向 |
| 作者 | 单位 | 邮编 | | 李汶华 | 茄子视频污app管理与经济学部 | 300072 | | 冯婧妮 | 茄子视频污app管理与经济学部 | | | 郭均鹏* | 茄子视频污app管理与经济学部 | 300072 |
|
| 中文摘要 |
| 现有研究多从用户角度出发，通过最大程度满足用户需求以实现平台效益最大化。然而，在长期背景下，仅从用户角度出发，内容提供者会因为没有足够的曝光而离开平台。考虑到用户关切的是平台对其推荐的内容与其偏好之间的相关性，而提供者关切的是平台对其曝光的公平性，本文将从平台策略角度出发，运用多智能体建模框架，分析平台对用户和提供者不同关切的权衡在长期背景下对个体和整体的影响，以探究平台的最优策略。首先，面对长期背景下提供者公平关切的问题，本文提出了长期和短期曝光在时间层面和用户层面的权衡，实现了曝光的平稳收敛；而后，面对提供者和用户两种不同的目标，设计不同偏好的策略，观察其长期表现。模拟结果表明，提供者曝光公平和用户偏好的矛盾并非完全不可调和，在内容供给固定的情况下，完全偏向用户反而会伤害用户信任，但适度偏向于用户的混合策略可以使平台的用户信任稳步上升，更有利于平台的长期发展。 |
| 英文摘要 |
| Previous research has largely concentrated on maximizing platform benefits by extensively meeting user demands. However, in the long term, exclusively adhering to the user perspective may result in provider attrition due to insufficient exposure on the platform. Recognizing that users prioritize the relevance of recommended content, while providers are concerned with exposure fairness, this article explores platform strategies through an Agent-Based Modelling framework. It analyses the trade-offs between meeting user preferences and ensuring fair exposure for providers over the long term to identify the optimal platform strategy. First, the study addresses the issue of provider fairness by proposing a trade-off between long-term and short-term exposure at temporal and user levels, ensuring a consistent convergence of exposure. Additionally, it designs various strategies tailored to user and provider preferences and evaluates performance longitudinally. Simulation suggest that the conflict between provider fairness and user preferences is not entirely irreconcilable. favouring users completely with a fixed content supply may undermine their trust. However, a mixed strategy with moderate user bias can steadily enhance user trust, ultimately benefiting the platform's long-term development. |
| 关闭 |
