R-GCN: The R Could Stand for Random

4 Mar 2022  ยท  Vic Degraeve, Gilles Vandewiele, Femke Ongenae, Sofie Van Hoecke ยท

The inception of the Relational Graph Convolutional Network (R-GCN) marked a milestone in the Semantic Web domain as a widely cited method that generalises end-to-end hierarchical representation learning to Knowledge Graphs (KGs). R-GCNs generate representations for nodes of interest by repeatedly aggregating parameterised, relation-specific transformations of their neighbours. However, in this paper, we argue that the the R-GCN's main contribution lies in this "message passing" paradigm, rather than the learned weights. To this end, we introduce the "Random Relational Graph Convolutional Network" (RR-GCN), which leaves all parameters untrained and thus constructs node embeddings by aggregating randomly transformed random representations from neighbours, i.e., with no learned parameters. We empirically show that RR-GCNs can compete with fully trained R-GCNs in both node classification and link prediction settings.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Node Classification AIFB RR-GCN-PPV Accuracy 86.11 # 6
Node Classification AIFB RR-GCN-PPV-CUT Accuracy 95.83 # 1
Node Classification AM RR-GCN-PPV-CUT (Unimportant relations removed) Accuracy 91.31 # 1
Node Classification AM RR-GCN-PPV-CUT Accuracy 84.8 # 6
Node Classification AM RR-GCN-PPV Accuracy 84.65 # 7
Node Classification AMPLUS RR-GCN-PPV Accuracy 84.54 # 1
Node Classification AMPLUS R-GCN Accuracy 83.81 # 2
Node Classification BGS RR-GCN-PPV Accuracy 78.97 # 6
Node Classification BGS RR-GCN-PPV-CUT Accuracy 84.14 # 4
Node Classification DBLP R-GCN Accuracy 68.51 # 3
Node Classification DBLP RR-GCN-PPV Accuracy 70.61 # 2
Node Classification DMG777K R-GCN Accuracy 62.51 # 2
Node Classification DMG777K RR-GCN-PPV Accuracy 63.97 # 1
Node Classification DMGFULL R-GCN Accuracy 57.52 # 2
Node Classification DMGFULL RR-GCN-PPV Accuracy 63.38 # 1
Link Prediction FB15k-237 RR-GCN-PPV MRR 0.238 # 62
Hits@10 0.412 # 65
Hits@3 0.256 # 46
Hits@1 0.157 # 51
Node Classification MDGENRE RR-GCN-PPV Accuracy 67.15 # 2
Node Classification MDGENRE R-GCN Accuracy 67.33 # 1
Node Classification MUTAG RR-GCN-PPV Accuracy 79.41 # 1

Methods


No methods listed for this paper. Add relevant methods here