摘要
为解决推荐系统中数据稀疏造成的用户冷启动问题,文中提出了一种基于方面级用户偏好迁移的跨领域推荐算法(Cross-Domain Recommendation via Review Aspect-Level User Preference Transfer, CAUT),设计了基于两阶段生成对抗网络的用户方面级偏好跨领域迁移结构,通过用户历史评论挖掘用户细粒度方面级偏好。CAUT利用预训练源领域编码器参数对目标领域编码器进行参数初始化,在固定源领域编码器参数的同时引入领域鉴别器,以解决源领域与目标领域数据分布差异的问题,进而可以有效利用源领域的丰富数据,缓解目标领域数据稀疏造成的用户冷启动问题。在亚马逊电商平台真实数据集上进行了实验,结果表明,与最新算法相比,CAUT在用户对商品的评分预测均方根误差(RMSE)指标上有明显的提升,说明CAUT可有效缓解用户冷启动问题。
In order to solve the user cold-start problem caused by data-sparse in recommender system, this paper proposes a cross-domain recommendation algorithm based on aspect-level user preference transfer, named CAUT.CAUT is devised to learn aspect transfer across domains from a two-stage generative adversarial network and extract aspect-level user fine-grained prefe-rence from reviews.The data distribution misalignment between source and target domains is eliminated by fixing source domain encoder parameters and designing a domain discriminator.Then the user cold-start problem caused by data-sparse in the target domain could be alleviated by utilizing source domain data via CAUT.Experiments on real-world datasets show that the proposed CAUT outperforms SOTA models significantly in rating prediction RMSE indicator, suggesting that CAUT can effectively solve the user cold-start problem.
作者
张佳
董守斌
ZHANG Jia;DONG Shou-bin(School of Computer Science and Engineering,South China University of Technology,Guangzhou 510006,China;Zhongshan Institute of Modern Industrial Technology of SCUT,Zhongshan,Guangdong 528437,China)
出处
《计算机科学》
CSCD
北大核心
2022年第9期41-47,共7页
Computer Science
关键词
跨领域推荐
方面级用户偏好
用户冷启动
生成对抗网络
Cross-domain recommendation
Aspect-level user preference
Cold-start user
Generative adversarial network
