用于冷启动推荐的用户偏好跨域转移框架  

User preference cross-domain transfer framework for cold-start recommendation

作　　者：刘玉芳 王绍卿[1] 郑顺 张丽杰 孙福振[1] LIU Yufang;WANG Shaoqing;ZHENG Shun;ZHANG Lijie;SUN Fuzhen(School of Computer Science and Technology,Shandong University of Technology,Zibo 255049,China)

机构地区：[1]山东理工大学计算机科学与技术学院,山东淄博255049

出　　处：《山东理工大学学报（自然科学版）》2024年第1期26-32,41,共8页Journal of Shandong University of Technology:Natural Science Edition

基　　金：山东省自然科学基金项目(ZR2020MF147,ZR2021MF017);山东省高等学校青创科技计划创新团队项目(2021KJ031)。

摘　　要：在推荐系统中冷启动问题一直面临着巨大的挑战。跨域推荐旨在利用源域中用户丰富的信息来提高目标域中冷启动用户推荐的性能,能够有效地缓解冷启动问题。基于映射函数的跨域推荐方法存在过于依赖重叠用户的问题,然而在冷启动场景中重叠用户的数量非常少,导致较差的泛化能力。元学习能够在冷启动推荐背景下快速适应数据稀疏的新任务,具有良好的泛化能力。提出了用于冷启动推荐的用户偏好跨域转移框架(UPCTFCR),设计了一个自编码器和元学习器,用以训练映射函数的初始化参数,将用户在源域中的偏好转移到目标域,将此作为冷启动用户在目标域中的初始嵌入进行推荐。利用亚马逊数据集构建了3个跨域任务,通过实验证明了UPCTFCR在冷启动推荐中的有效性。The problem of cold-start in recommendation system has been a great challenge.Cross-domain recommendation aims to improve the recommendation performance for cold-start user in the target domain by utilizing the rich user information in the source domain and effectively alleviates the cold-start problem.Cross-domain recommendation method based on mapping function relies too much on overlapping users.However,the number of overlapping users in the cold-start scenario is very small,resulting in poor generalization ability.Meta-learning can quickly adapt to new tasks with sparse data in the context of cold-start recommendation,and had good generalization ability.This paper proposes a user preference cross-domain transfer framework for cold-start recommendation(UPCTFCR).An autoencoder and a meta-learner were designed to train the initialization parameters of the mapping function and transferred the preferences of the user in the source domain to the target domain as the initial embedding of the cold-start user recommendation.Three cross-domain tasks were constructed using the Amazon dataset,and experiments were conducted to demonstrate the effectiveness of UPCTFCR in cold-start recommendations.

关 键 词：跨域推荐 冷启动推荐 元学习 自注意力 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]