基于多臂老虎机的异质网络表示学习方法
Heterogeneous network representation learning based on multi-armed bandit
闫旸 1陈泽秋 1邓钧霖1
作者信息
- 1. 天津职业技术师范大学信息技术工程学院,天津 300222
- 折叠
摘要
针对异质网络表示学习中邻接节点表示向量的融合问题,提出基于多臂老虎机的异质网络表示学习方法.该方法采用基于多臂老虎机思想,实现异质网络中元路径关系的权重的自适应计算,在节点分类任务上取得的Micro-F1值(89.56%和54.79%)和Macro-F1值(89.09%和53.14%)均优于基准测试.面对节点信息的多样性,基于多臂老虎机的网络表示学习方法能够将网络结构和节点信息更加有效地融入图的表示学习中.
Abstract
To deal with the fusion of adjacent node representation vectors in heterogeneous network representation learn-ing,an adaptive calculation of relationship weights of meta-path relations in heterogeneous networks is achieved by adopting the idea of multi-armed bandit.By introducing the above measures,this method achieved Micro-F1 values of 89.56%and 54.79%,as well as Macro-F1 values of 89.09%and 53.14%in node classification tasks.The research results show that in the face of the diversity of node information,the network representation learning method based on multi-armed bandit can more effectively integrate network structure and node information into graph representation learning.
关键词
多臂老虎机模型/异质网络/网络表示学习/自适应权重Key words
muti-arm bandit/heterogeneous network/network representation/self-adaptive weight引用本文复制引用
基金项目
教育部人文社会科学研究规划基金青年基金(22YJC870018)
天津市教委科研项目(2020KJ112)
应用数学福建省高等学校重点实验室开放基金(SX201904)
天津职业技术师范大学人才启动项目(KYQD1817)
出版年
2024
NETL
NSTL
万方数据