Hierarchical graph attention networks for semi-supervised node classification
- PDF / 2,908,054 Bytes
- 11 Pages / 595.224 x 790.955 pts Page_size
- 106 Downloads / 210 Views
Hierarchical graph attention networks for semi-supervised node classification Yixiong Feng1 · Kangjie Li1 · Yicong Gao1 · Jian Qiu2
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Recently, there has been a promising tendency to generalize convolutional neural networks (CNNs) to graph domain. However, most of the methods cannot obtain adequate global information due to their shallow structures. In this paper, we address this challenge by proposing a hierarchical graph attention network (HGAT) for semi-supervised node classification. This network employs a hierarchical mechanism for the learning of node features. Thus, more information can be effectively obtained of the node features by iteratively using coarsening and refining operations on different hierarchical levels. Moreover, HGAT combines with the attention mechanism in the input and prediction layer. It can assign different weights to different nodes in a neighborhood, which helps to improve accuracy. Experiment results demonstrate that state-of-theart performance was achieved by our method, not only on Cora, Citeseer, and Pubmed citation datasets, but also on the simplified NELL knowledge graph dataset. The sensitive analysis further verifies that HGAT can capture global structure information by increasing the receptive field, as well as the effective transfer of node features. Keywords Graph convolutional networks · Hierarchical representation · Semi-supervisded
1 Introduction Graphs can encode complex geometric structures that lie in the non-Euclidian domain. They can be studied with strong mathematical tools [1], and nowadays have become ubiquitous. For example, in e-commerce, to make accurate recommendations, it is necessary to exploit the interactions between users and products [2, 3]. In chemistry, a new drug can be discovered by using a graph-based learning method that models the molecules as a graph [4]. In citation Yixiong Feng
[email protected] Kangjie Li [email protected] Yicong Gao [email protected] Jian Qiu [email protected] 1
State Key Laboratory of Fluid Power and Mechatronic Systems, Zhejiang University, Hangzhou, 310027, China
2
School of Cyberspace, Hangzhou Dianzi University, Hangzhou, 310027, China
networks, papers can be categorized into different groups through their citation graphs [5, 6]. Moreover, there are many unlabeled data in the real world and labeling data is sometimes unrealistic and time-consuming. The semisupervised manner means that only a small amount of labeled data is used to train the model. Consequently, it is often crucial to analyze graphs in that situation, and the key issue is to maximize the effective utilization of the feature information of the unlabeled data [7]. As an approach to graph analysis, graph neural networks (GNNs) are closely related to graph embedding. Graph embedding [8] is a method that learns to represent graph nodes in low-dimensional vectors. Approaches such as word embedding [9], DeepWalk [10], node2vec [11] have achieved a breakthrough. However, be
Data Loading...
