.png){kind=icon}
返回
HHGNN: Hyperbolic Hypergraph Convolutional Neural Network based on variational autoencoder
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2024.128225.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
In recent years, there has been a growing interest in the widespread application of graph neural networks (GNNs). However, existing GNN frameworks are predominantly designed for simple graphs in Euclidean space, limiting their effectiveness in handling scale-free graph-structured data that is multimodal and multiscale. Recently, there has been a surge in approaches using hyperbolic spaces to better model scale-free graphs and overcome the limitations of Euclidean space. Nevertheless, these methods face challenges in effectively handling hierarchical multimodal data. To address this gap and leverage multilevel aggregation for capturing high-order hidden information in local representations, we propose the Hyperbolic Hypergraph Convolutional Neural Network (HHGNN). This deep graph representation learning framework, based on Variational Autoencoder (VAE), maps scale-free graphs from Euclidean space to hyperbolic space. In HHGNN, we define the hypergraph convolutional neural networks in hyperbolic space. Furthermore, considering the multimodal nature of data representations, our model demonstrates strong scalability, currently supporting over ten data formats and capable of constructing hypergraph inputs for HHGNN training. Extensive benchmark experiments demonstrate the outstanding performance of HHGNN in node classification tasks, particularly on datasets with hierarchical structures, outperforming current methods. The experimental results also illustrate that our model effectively captures data distribution characteristics and enhances data representation capabilities. Additionally, we conduct an in-depth analysis of a diabetes dataset, aiming to support early diabetes diagnosis.
Keyword:
Hyperbolic hypergraph
Multimodal data
Graph convolutional neural network
Variational autoencoder
期刊
IF:
6.5
论文数:
2.5W
被引数:
6.5W
机构
引用论文
Neighbor-aware deep multi-view clustering via graph convolutional network.png)
INFORMATION FUSION
INFORMATION FUSIONIF15.5
PREAI
分享收藏In Situ Monitoring of the Aging of Mineral Particles by Methylglyoxal Using Infrared Spectroscopy. Part I: Unraveling Uptake under Dry Conditions利用甲基乙二醛通过红外光谱法原位监测矿物颗粒老化。第一部分：解析干燥条件下的吸收机理。
PREAI
分享收藏Enabling Homogeneous GNNs to Handle Heterogeneous Graphs via Relation Embedding使同构gnn能够通过关系嵌入处理异构图
OAAI
OAAI分享收藏
查看更多内容