FRC:一种基于图嵌入对抗学习公平表示的推荐算法  

FRC:A Recommendation Algorithm Based on Graph Embedding Adversarial Learning for Fair Representation

作　　者：李昊洋 Haoyang Li(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海

出　　处：《建模与仿真》2024年第3期3515-3524,共10页Modeling and Simulation

基　　金：国家自然科学基金项目(61772342)资助。

摘　　要：推荐系统的目标是过滤和筛选信息,从而预测用户最感兴趣的内容。但公平性问题可能会对推荐结果产生影响。当前的公平性推荐模型大多数假设用户实体之间的独立性,没有考虑实体之间相关的情况下该如何缓解公平性问题。此外,在已有的少数利用图结构缓解公平性问题的研究中,仅利用节点嵌入学习数据表示,而用户的敏感信息可能会被局部图结构暴露。针对上述两个问题,本文提出了一种基于图嵌入对抗学习实现公平推荐的模型FRC,适用于任何将用户和项目嵌入作为输入的推荐任务,该模型将用户–项目数据映射为双向图,结合用户级和节点级的嵌入表示,利用对抗学习消除敏感特征以获得推荐任务中的公平表示。在两个真实世界数据集上的对比实验结果表明,本文提出的模型在推荐任务中为用户生成了更公平的推荐效果。The goal of recommendation systems is to filter and select information in order to predict the content that users are most interested in.However,fairness issues may impact the recommendation results.Most existing fairness recommendation models assume independence between user entities and do not consider how to mitigate fairness issues when entities are correlated.Additionally,in a few existing studies that utilize graph structures to address fairness issues,only node embeddings are used for learning data representations,potentially exposing sensitive user information through local graph structures.To address these two issues,this paper proposes a model called FRC,based on graph embedding adversarial learning,to achieve fair recommendations.It is applicable to any recommendation task that takes user and item embeddings as input.The model maps user-item data into a bidirectional graph,combines user-level and node-level embeddings,and utilizes adversarial learning to eliminate sensitive features for fair representation in recommendation tasks.Comparative experiments on two real-world datasets demonstrate that the proposed model generates fairer recommendations for users in recommendation tasks.

关 键 词：对抗学习 公平表示学习 图嵌入 推荐系统 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]