Large-scale multi-view attributed graph datasets: MAG-eng & MAG-phy

The large-scale MAG-eng and MAG-phy datasets are released in a zip file consisting of corresponding data files. Each multi-view attributed graph is stored in a pickle file, e.g., new_mageng.pkl. 

Usage: 

cd /SGLA
unzip /mag_data.zip -d data/

Data specification:

Key
Description

'PA'
The sparse matrix G that encodes Publication-Author relationships. The co-author graph view G_1 is acquired via G@G.T.

'PP'
The adjacency matrix of citation relationships, i.e., the citation graph view G_2.

'features'
The first attribute view, extracted from the publication abstracts.

'features1'
The second attribute view, extracted from field-of-study terms.

'label'
Cluster/class label of publication nodes.