CoMGL: Collaborative Multi-view Graph Learning for Strict Cold-start Tasks

Data

Data Preparation for OGB-MAG dataset.

• Main view: ("paper", "to", "author")
• Auxiliary views: ("paper", "to", "Conference"), ("paper", "to", "term")

Data Process

• Divide the train/val/test dataset, remove the redundant edges of each part of the dataset, but keep all the edges of the auxiliary view to satisfy the inductive setting;
• Remove all edges of the main view in the val/test dataset to meet the "Strict Cold Start" scenario.

python preprocess.py --data_path data/

Experiments