面向冷启动用户的元学习与图转移学习序列推荐  

Sequential recommendation for cold-start users with meta graph transitional learning

作　　者：李璐 张志军[1] 范钰敏 王星 袁卫华[1] LI Lu;ZHANG Zhijun;FAN Yumin;WANG Xing;YUAN Weihua(School of Computer Science and Technology,Shandong Jianzhu University,Jinan 250101,Shandong,China)

机构地区：[1]山东建筑大学计算机科学与技术学院,山东济南250101

出　　处：《山东大学学报（工学版）》2024年第2期69-79,共11页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(61902221,62177031);山东省自然科学基金资助项目(ZR2021MF099,ZR2022MF334);山东省教学改革研究项目(M2021130,M2022245,Z2022202);山东省优质专业学位教学案例库建设项目(SDYAL2022155);山东省重点研发计划(软科学项目)资助项目(2021RKY03056)。

摘　　要：为解决推荐系统用户冷启动问题,提出面向冷启动用户的元学习与图转移学习序列推荐(sequential recommendation for cold-start users with meta graph transitional learning, MetaGTL)。MetaGTL在不使用其他辅助信息的前提下,采用图神经网络(graph neural network, GNN)建模序列间物品高阶关系生成用户物品嵌入;将交互序列构造为物品对集合,使用序列编码模块捕捉物品间的转移关系,动态建模用户兴趣;采用注意力机制,生成准确的用户特征;采用基于梯度的元学习方法训练模型,生成初始化模型;对模型的工作性能和结果进行详细分析,结合基线模型进行对比评价。试验结果表明,基于元学习与图转移学习的MetaGTL在缺少辅助信息的用户冷启动任务中具有更高的预测精度。To solve the cold-start problem for users of the recommendation systems,a sequential recommendation for cold-start users with meta graph transitional learning(MetaGTL)was proposed.MetaGTL used a graph neural network(GNN)to model higher order relationship between sequences of items to generate user embeedings and item embeddings without the use of auxiliary information.The sequence encoder constructed interacted sequences as sets of item pairs and captured transferred relationship among items.The user interest representation module used the attention mechanism to generate accurate user profile.The gradient-based meta-learning method was used to train the model to obtain an initialization model.The performance and result of the proposed approach were analyzed in detail through comparing with the baseline models.The experimental results showed that compared with other model methods,the proposed MetaGTL-based on meta-learning and graph transfer learning had higher prediction accuracy in user cold-start tasks without auxiliary information.

关 键 词：推荐系统 序列推荐 用户冷启动 图神经网络 元学习 深度学习 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]