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
李璐 1张志军 1范钰敏 1王星 1袁卫华1
作者信息
- 1. 山东建筑大学计算机科学与技术学院,山东济南 250101
- 折叠
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.
Abstract
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 or-der relationship between sequences of items to generate user embeedings and item embeddings without the use of auxiliary informa-tion.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 mod-el 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Key words
recommendation system/sequence recommendation/user cold-start/graph neural network/meta-learning/deep learning引用本文复制引用
基金项目
国家自然科学基金(61902221)
国家自然科学基金(62177031)
山东省自然科学基金(ZR2021MF099)
山东省自然科学基金(ZR2022MF334)
山东省教学改革研究项目(M2021130)
山东省教学改革研究项目(M2022245)
山东省教学改革研究项目(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(软科学研究计划)资助项目(2021RKY03056)
出版年
2024
NETL
NSTL
万方数据