TY  - JOUR
T1  - Graph Convolutional Networks with Elaborate Neighborhood Selection
AU  - Jung, Yeonsung 
AU  - Whang, Joyce Jiyoung 
JO  - Journal of KIISE, JOK
PY  - 2019
DA  - 2019/1/14
DO  - 10.5626/JOK.2019.46.11.1193
KW  - node classification
KW  - graph convolutional network
KW  - graph neural network
KW  - graph mining
AB  - Graph Convolutional Networks (GCNs) utilize the convolutional structure to obtain an effective insight on representation by aggregating the information from neighborhoods. In order to demonstrate high performance, it is necessary to select neighborhoods that can propagate important information to target nodes, and acquire appropriate filter values during training. Recent GCNs algorithms adopt simple neighborhood selection methods, such as taking all 1-hop nodes. In the present case, unnecessary information was propagated to the target node, resulting in degradation of the performance of the model. In this paper, we propose a GCN algorithm that utilizes valid neighborhoods by calculating the similarity between the target node and neighborhoods.