GIDN: A Lightweight Graph Inception Diffusion Network for High-efficient Link Prediction

4 Oct 2022  ·  Zixiao Wang, Yuluo Guo, Jin Zhao, Yu Zhang, Hui Yu, Xiaofei Liao, Hai Jin, Biao Wang, Ting Yu ·

In this paper, we propose a Graph Inception Diffusion Networks(GIDN) model. This model generalizes graph diffusion in different feature spaces, and uses the inception module to avoid the large amount of computations caused by complex network structures. We evaluate GIDN model on Open Graph Benchmark(OGB) datasets, reached an 11% higher performance than AGDN on ogbl-collab dataset.

PDF Abstract
No code implementations yet. Submit your code now


Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Link Property Prediction ogbl-collab GIDN@YITU Test Hits@50 0.7096 ± 0.0055 # 1
Validation Hits@50 0.9620 ± 0.0040 # 8
Number of params 60449025 # 3
Ext. data No # 1
Link Property Prediction ogbl-ddi GIDN@YITU Test Hits@20 0.9542 ± 0.0000 # 1
Validation Hits@20 0.8258 ± 0.0000 # 5
Number of params 3506691 # 5
Ext. data No # 1