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

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。

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.