by WangYC @NWPU changan Apr.20th 2022

ps:only using the DGI_GCN parts for the baseline.

pytorch==0.5 or 0.4 python==3.8

cd HDGI-master/DGI_HGCN/ python execute.py dataset patience

Parameter Setting:

Node-level dimension: 16
Attention head: 4
Semantic-level attention vector: 8
learning rate: 0.02

Node-level dimension: 64
Semantic-level attention vector: 8
learning rate: 0.02

Node-level dimension: 16
Attention head: 4
learning rate: 0.005
Drop out ratio: 0.6

Hidden-unit dimension: 64
learning rate: 0.01
Drop out ratio: 0.5

Hidden-unit dimension: 16
learning rate: 0.01
Drop out ratio: 0

Embedding dimension: 100
learning rate: 0.01
negative-samples: 5
Window: 1

Node-level dimension: 16
Attention head: 4
Semantic-level dimension: 8
learning rate: 0.005
Node-level Drop out ratio: 0.6
Semantic-level Drop out ratio: 0.6

Number of walks per node: 10
Dimensions of word embeddings: 128
Length of random walk: 30
Window size for skipgram: 5

Hidden-unit dimension: 64
learning rate: 0.001
Drop out ratio: 0