行为增强的多层次协同Top-N推荐
Multilevel collaborative top-n recommendation based on enhanced behavior
刘宇鹏 1吕衍河1
作者信息
- 1. 哈尔滨理工大学 计算机科学与技术学院,黑龙江 哈尔滨 150001
- 折叠
摘要
传统推荐系统只利用单一用户行为,然而用户行为间是具有关联性,忽视用户行为会丢失辅助行为对目标行为的影响.本文提出了一种行为增强的多层次协同Top-N推荐,在推荐二分图和元路径图上利用注意力机制传播信息,学习多层次高阶和异质协同信号(包括用户-项目间的和项目间的)以提高推荐性能,这样可以更好地利用推荐图结构,并充分考虑到推荐图结构上各种行为间的相互影响.在经典数据集上做了全方位实验验证模型有效性,在电商推荐数据上取得了很好效果.
Abstract
Traditional recommendation systems only use one type of user behavior.However,multiple behaviors of users are related;therefore,ignoring user behaviors will result in the loss of the influence of auxiliary behavior on the target behavior.This paper proposes a multilevel collaborative Top-N recommendation based on enhanced user behavior(MCREB),which uses the attention mechanism to propagate information on the recommendation bipartite and item-based metapath graphs and learns multilevel high-order and heterogeneous collaborative signals,including user-item and inter-item,to improve recommendation performance.Thus,the model can better use the recommen-dation graph structure and fully consider the interaction between various behaviors on the recommendation graph structure.Furthermore,comprehensive experiments are conducted on the benchmark dataset to verify the model's effectiveness.
关键词
辅助行为/多行为/图神经网络/元路径图/用户-项目/传播层/目标行为/高阶异质信号Key words
auxiliary behavior/multibehavior/graph neural network/metapath graph/user-item/propagation lay-er/target behavior/high-order heterogenous signal引用本文复制引用
基金项目
国家自然科学基金(61300115)
中国博士后科学基金(2014m561331)
黑龙江省教育厅科学技术研究项目(12521073)
出版年
2024
国家科技期刊平台
NSTL
万方数据