Open AccessAcademic Network Representation Learning Based on Metapath Tree
Author(s) -
Wei Zhang,
Ying Liang,
Xiangxiang Dong
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1302/2/022077
Subject(s) - computer science , representation (politics) , machine learning , graph , artificial intelligence , cluster analysis , process (computing) , tree (set theory) , variety (cybernetics) , sampling (signal processing) , feature learning , theoretical computer science , data mining , mathematics , mathematical analysis , filter (signal processing) , politics , political science , law , computer vision , operating system
Network representation learning aims to use low-dimensional dense vectors to represent nodes in the graph, which can reflect the graph structure and can be used in a variety of machine learning tasks. The academic network contains richer information, which most of the current methods are unable to capture. This paper proposes a method to get better vectors in the academic network. This method first uses the metapath tree to guide the random walk process, and adds a sampling process to preserve multiple metapath information. The vector representation of the nodes is obtained by training using the skip-gram model on the academic network. Experiment results show that the proposed model outperforms several traditional network representation learning models in multi-label classification and clustering tasks.