TY  - JOUR
T1  - Graph Node Importance Estimation Tasks : A Review of Models and Analyzing Importance of Topology
AU  - Cho, Woo-seong 
AU  - Song, Seung-heon 
AU  - Lee, Jae-koo 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2024
DA  - 2024/1/1
DO  - 10.7840/kics.2024.49.5.734
KW  - Node Importance Estimation
KW  - Graph
KW  - Graph Neural Network
KW  - Inductive Learning
AB  - The importance of nodes in graph data is utilized across various fields. It is used for selecting products for user recommendations in OTT (Over-The-Top) platforms and e-commerce services, determining priority in query searches within knowledge graphs, and prioritizing the allocation of limited network resources. This paper summarizes and compares node importance estimation models, considering the training method and the scope of data used, including input importance, edge types, and node centrality. The models covered in this paper are limited to supervised learning methods. So, for situations where labels are unavailable or sparse, we conduct inductive learning with finetuning to use the models. Additionally, as the node feature vectors are defined differently between source and target data, we also train and test the models without node feature vectors to investigate their impact. The results show that overall performance degradation was generally not significant. This highlights the importance of graph topology in node importance estimation, suggesting that high-quality node importance can be obtained through inductive learning even in situations where labels are unavailable or sparse. The insights from this research can greatly contribute to studies that require node importance estimation in various fields.