Journal
PROCEEDINGS OF THE 30TH ACM INTERNATIONAL CONFERENCE ON INFORMATION & KNOWLEDGE MANAGEMENT, CIKM 2021
Volume -, Issue -, Pages 2827-2831Publisher
ASSOC COMPUTING MACHINERY
DOI: 10.1145/3459637.3482215
Keywords
Link Prediction; Graph Embedding; Graph Convolutional Networks; Normalization
Categories
Funding
- Bio-Synergy Research Project of the MSIT [2013M3A9C4078137]
- National Research Foundation of Korea(NRF) [2020R1A2C1004032]
- ITRC support program [IITP-2020-2020-0-01795]
Ask authors/readers for more resources
Link prediction is a key problem for graph-structured data, with graph autoencoders and variational graph autoencoders proposed to learn graph embeddings. However, existing methods do not perform well in link predictions involving isolated nodes. A new VGNAE model has been introduced, utilizing L-2 normalization for better embeddings of isolated nodes.
Link prediction is one of the key problems for graph-structured data. With the advancement of graph neural networks, graph autoencoders (GAEs) and variational graph autoencoders (VGAEs) have been proposed to learn graph embeddings in an unsupervised way. It has been shown that these methods are effective for link prediction tasks. However, they do not work well in link predictions when a node whose degree is zero (i.g., isolated node) is involved. We have found that GAEs/VGAEs make embeddings of isolated nodes close to zero regardless of their content features. In this paper, we propose a novel Variational Graph Normalized AutoEncoder (VGNAE) that utilize L-2-normalization to derive better embeddings for isolated nodes. We show that our VGNAEs outperform the existing state-of-the-art models for link prediction tasks. The code is available at https://github.com/SeongJinAhn/VGNAE.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available