RHGNN

RHGNN

Paper:Heterogeneous Graph Representation Learning with Relation Awareness

Code from author:https://github.com/yule-BUAA/R-HGNN

How to run

Clone the Openhgnn-DGL

python main.py -m RHGNN -t node_classification -d imdb4GTN -g 0 --use_best_config

Candidate dataset: imdb4GTN/acm4GTN

If you do not have gpu, set -gpu -1.

candidate dataset

imdb4GTN

NOTE: imdb4GTN is a small-scale dataset. We will add large-scale datasets (OGB-MAG, OAG-Venue and OAG-L1-Field) in our further work.

performance

Node classification

	accuracy
imdb4GTN	0.5883

TrainerFlow: node_classification

model

• RHGNN

  • Contain two R_HGNN_layer、 relation_fusing and one classifier

• R_HGNN_layer

  • first step: hetero_conv and residual connection

    • HeteroGraphConv
    • Use RelationGraphConv to create hetero_conv, each RelationGraphConv deals with a single type of relation
    • A generic module for computing convolution on heterogeneous graphs

  • second step: relation_crossing_layer

    • RelationCrossing
    • Create relation_crossing_layer
    • Establish connections of node representations to improve message passing across different relations and automatically distinguish the importance of relations

  • third step: relation_propagation_layer

  R-HGNN layer is composed of these three components and stack L layers to receive information from multi-hop neighbors. Finally, the L layers could provide relation-aware node representations for target node.

• RelationFusing

  • Aggregate the relation-aware node representations into a compact node representation
  • Create relation_fusing

Hyper-parameter specific to the model

num_heads  = 8            
hidden_units = 64
relation_hidden_units = 8
n_layers = 2
learning_rate = 0.001
dropout = 0.5
residual = True

Best config can be found in best_config

More

Contirbutor

Tianyu Zhao, Qi Zhang[GAMMA LAB]

If you have any questions,

Submit an issue or email to [email protected].