《计算机应用》唯一官方网站 ›› 2021, Vol. 41 ›› Issue (12): 3508-3514.DOI: 10.11772/j.issn.1001-9081.2021060910

• 第十八届中国机器学习会议(CCML 2021) • 上一篇    

基于改进的倾向得分估计的无偏推荐模型

骆锦潍1,2,3, 刘杜钢1,2,3, 潘微科1,2,3(), 明仲1,2,3   

  1. 1.大数据系统计算技术国家工程实验室(深圳大学),广东 深圳 518060
    2.人工智能与数字经济广东省实验室(深圳)(深圳大学),广东 深圳 518060
    3.深圳大学 计算机与软件学院,广东 深圳 518060
  • 收稿日期:2021-05-12 修回日期:2021-06-22 接受日期:2021-06-29 发布日期:2021-12-28 出版日期:2021-12-10
  • 通讯作者: 潘微科
  • 作者简介:骆锦潍(1997—),男,广东汕尾人,硕士研究生,主要研究方向:推荐系统、迁移学习、反事实学习
    刘杜钢(1993—),男,福建泉州人,博士研究生,主要研究方向:推荐系统、因果推断
    明仲(1967—),男,江西宁都人,教授,博士,CCF高级会员,主要研究方向:人工智能、网络智能。
  • 基金资助:
    国家自然科学基金重点项目(61836005);国家自然科学基金面上项目(61872249)

Unbiased recommendation model based on improved propensity score estimation

Jinwei LUO1,2,3, Dugang LIU1,2,3, Weike PAN1,2,3(), Zhong MING1,2,3   

  1. 1.National Engineering Laboratory for Big Data System Computing Technology (Shenzhen University),Shenzhen Guangdong 518060,China
    2.Guangdong Laboratory for Artificial Intelligence and Digital Economy (Shenzhen) (Shenzhen University),Shenzhen Guangdong 518060,China
    3.College of Computer Science and Software Engineering,Shenzhen University,Shenzhen Guangdong 518060,China
  • Received:2021-05-12 Revised:2021-06-22 Accepted:2021-06-29 Online:2021-12-28 Published:2021-12-10
  • Contact: Weike PAN
  • About author:LUO Jinwei, born in 1997, M. S. candidate. His research interests include recommender system, transfer learning, counterfactual learning.
    LIU Dugang, born in 1993, Ph. D. candidate. His research interests include recommender system, causal inference.
    MING Zhong, born in 1967, Ph. D., professor. His research interests include artificial intelligence, Web intelligence.
  • Supported by:
    the Key Program of National Natural Science Foundation of China(61836005);the Surface Program of National Natural Science Foundation of China(61872249)

摘要:

现实中推荐系统通常遭受着各种各样的偏置问题,例如曝光偏置、位置偏置和选择偏置。一个忽略偏置问题的推荐模型不能反映推荐系统的真实性能,且对于用户而言可能是不可信任的。先前的工作已经表明基于倾向得分估计的推荐模型能够有效缓解隐式反馈数据的曝光偏置,但是通常只考虑通过物品信息来估计倾向得分,这可能导致倾向得分估计不准确。为了提高倾向得分估计的准确性,提出配对倾向得分估计(MPE)方法。具体来说,该方法引入了用户流行度偏好的概念,通过计算用户流行度偏好和物品流行度的配对程度来对样本曝光率进行更加精确的建模,最后将提出的估计方法和一个主流的传统推荐模型以及一个无偏推荐模型进行集成并和包括前两者的三个基线模型进行对比。在公开数据集上的实验结果表明,结合MPE方法后的模型分别相比对应的基线模型在召回率、折损累计增益(DCG)和平均准确率(MAP)这三个评估指标上均有显著的提升;此外,通过实验结果还观察到性能的增益有很大一部分来自长尾物品,可见所提方法有助于提升推荐物品的多样性与覆盖率。

关键词: 推荐系统, 隐式反馈, 曝光偏置, 倾向得分估计, 矩阵分解, 长尾物品, 用户流行度偏好

Abstract:

In reality, recommender systems usually suffer from various bias problems, such as exposure bias, position bias and selection bias. A recommendation model that ignores the bias problems cannot reflect the real performance of the recommender system, and may be untrustworthy for users. Previous works show that a recommendation model based on propensity score estimation can effectively alleviate the exposure bias problem of implicit feedback data in recommender systems, but only item information is usually considered to estimate propensity scores, which may lead to inaccurate estimation of propensity scores. To improve the accuracy of propensity score estimation, a Match Propensity Estimator (MPE) method was proposed. Specifically, a concept of users’ popularity preference was introduced at first, and then more accurate modeling of the sample exposure rate was achieved by calculating the matching degree of the user’s popularity preference and the item’s popularity. The proposed estimation method was integrated with a traditional recommendation model and an unbiased recommendation model, and the integrated models were compared to three baseline models including the above two models. Experimental results on a public dataset show that the models combining MPE method achieve significant improvement on three evaluation metrics such as recall, Discounted Cumulative Gain (DCG) and Mean Average Precision (MAP) compared with the corresponding baseline models respectively. In addition, experimental results demonstrate that a large part of the performance gain comes from long-tail items, showing that the proposed method is helpful to improve the diversity and coverage of recommended items.

Key words: recommender system, implicit feedback, exposure bias, propensity score estimation, matrix factorization, long-tail item, users’ popularity preference

中图分类号: