基于跨域元学习框架的冷启动用户表示学习方法  

Cold-start user representation learning method based on cross-domain meta-learning framework

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 255000,Shandong,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255000

出　　处：《山东大学学报(工学版)》2024年第6期29-37,共9页Journal of Shandong University(Engineering Science)

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

摘　　要：为解决跨域推荐方法过度依赖重叠用户、在冷启动场景中由于数据稀疏导致泛化能力差两个问题,利用元学习快速适应数据稀疏任务的优势,提出一个基于跨域元学习框架的冷启动用户表示学习方法。设计一个多级注意力融合机制,门控循环单元(gate recurrent unit, GRU)获取用户的短期偏好,多级特征注意力融合源域中用户的长短期偏好,获取用户的广义表示。设计一个元网络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,获得冷启动用户在目标域中的初始嵌入表示,并以此进行推荐,取得较好结果。利用亚马逊数据集构建了3个跨域推荐任务并进行广泛试验,试验结果表明,本研究模型在平均绝对误差和均方根误差评价中均优于其他基线模型。To solve the two problems of over-reliance on overlapping users and poor generalization ability due to data sparsity in cold-start scenarios,which existed in cross-domain recommendation methods,and took advantage of meta-learning′s ability to quickly adapt to data-sparse tasks,a cold-start user representation learning method based on a cross-domain meta-learning framework was proposed.A multi-level attention fusion mechanism was first designed,where the gate recurrent unit extracted the user′s short-term preferences and the multi-level feature attention fused the user′s long short-term preferences in the source domain to obtain the user′s generalized representation.A meta-network was designed to train the initialization parameters of the mapping function to transfer the user′s preferences in the source domain to the target domain to obtain the initial embedded representation of the cold-start user in the target domain and used it to make a recommendation to achieve better results.Three cross-domain recommendation tasks were constructed using the Amazon dataset,and extensive experiments were conducted,the test results indicated that the model in this study outperformed other baseline models in terms of both mean absolute error and root mean square error evaluations.

关 键 词：跨域推荐 冷启动推荐 元学习 长短期偏好 多级注意力 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]