图书情报工作 ›› 2021, Vol. 65 ›› Issue (12): 101-111.DOI: 10.13266/j.issn.0252-3116.2021.12.010

• 研究论文 • 上一篇    下一篇

面向用户长短期偏好调节的可解释个性化推荐方法研究

李伟卿1,2, 池毛毛1, 王伟军2   

  1. 1. 华中师范大学信息管理学院 武汉 430079;
    2. 华中师范大学青少年网络心理与行为教育部重点实验室 武汉 430079
  • 收稿日期:2020-12-09 修回日期:2021-02-24 出版日期:2021-06-20 发布日期:2021-07-03
  • 通讯作者: 王伟军(ORCID:0000-0002-6077-1101),教授,博士,博士生导师,通讯作者,E-mail:wangwj@ccnu.edu.cn
  • 作者简介:李伟卿(ORCID:0000-0002-4108-2829),博士研究生;池毛毛(ORCID:0000-0003-2726-5933),副教授,博士。
  • 基金资助:
    本文系国家自然科学基金项目"面向青少年网络适应的个性化信息服务优化方法研究"(项目编号:71974072)和国家自然科学基金项目"基于屏幕视觉热区的网络用户偏好提取及交互式个性化推荐研究"(项目编号:71571084)研究成果之一。

Explainable Personalized Recommendation Method Based on Adjustment of Users' Long- and Short-Term Preferences

Li Weiqing1,2, Chi Maomao1, Wang Weijun2   

  1. 1. School of Information Management, Central China Normal University, Wuhan 430079;
    2. Key Laboratory of Adolescent Cyber Psychology and Behavior, Ministry of Education, Wuhan 430079
  • Received:2020-12-09 Revised:2021-02-24 Online:2021-06-20 Published:2021-07-03

摘要: [目的/意义] 针对目前推荐模型愈加复杂、数据输入越来越多、传统推荐模型可解释性较低、推荐结果"过度特化"等问题,提出面向用户长短偏好调节的可解释个性化推荐方法。[方法/过程] 从用户近期产品需求及其长期生活方式两个维度构建用户长短偏好模型,借鉴用户评分偏置及注意力机制,将用户长短偏好与其评分相结合进行评分预测,从而形成Top-N推荐。[结果/结论] 通过在两个数据集上的实验结果表明,本方法对于不同的用户行为(显式反馈或隐式反馈),不同的推荐项目个数及在不同的推荐算法中都有良好表现。在无需对各种推荐模型进行较大改变的情况下,提升了推荐结果的准确率、召回率与多样性;另外基于对长短偏好系数的改变,实现对推荐结果多样性与准确率的调整,并且形成相应的推荐解释。

关键词: 个性化推荐, 用户偏好, 可解释性, 长短偏好调节

Abstract: [Purpose/significance] We put forward an explainable personalized recommendation method based on adjustment of users' long- and short-preferences in view of the current problems that increasingly complexity and more feature data inputs of recommendation models, low interpretability of traditional recommendation models and over-specialization of recommendation results.[Method/process] We constructed a user preference model from two dimensions of users' recent product needs and their long-term lifestyles, used the user's rating bias and attention mechanism for reference, combined the user's long- and short-term preference with their direct score to predict the score of unknown items, and formed the Top-N recommendations.[Result/conclusion] The experimental results on two datasets showed that our method had a good performance to different user behaviors (explicit feedback or implicit feedback), different number of Top-N recommended items, and in different recommendation algorithms. It improves the accuracy, recall and diversity of the recommendation results without making great changes to various recommendation models, and based on the change of long- and short-term preference coefficients, it realizes the adjustment to the diversity and accuracy of the recommendation results, and form the corresponding recommended explanation.

Key words: personalized recommendation, user preference, explainable recommendation, adjustment of long- and short-preference

中图分类号: