Vic Degraeve, Gilles Vandewiele, Femke Ongenae, Sofie Van Hoecke
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.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Link Prediction | FB15k-237 | Hits@1 | 0.157 | RR-GCN-PPV |
| Link Prediction | FB15k-237 | Hits@10 | 0.412 | RR-GCN-PPV |
| Link Prediction | FB15k-237 | Hits@3 | 0.256 | RR-GCN-PPV |
| Link Prediction | FB15k-237 | MRR | 0.238 | RR-GCN-PPV |
| Node Classification | AMPLUS | Accuracy | 84.54 | RR-GCN-PPV |
| Node Classification | AMPLUS | Accuracy | 83.81 | R-GCN |
| Node Classification | DBLP | Accuracy | 70.61 | RR-GCN-PPV |
| Node Classification | DBLP | Accuracy | 68.51 | R-GCN |
| Node Classification | AIFB | Accuracy | 95.83 | RR-GCN-PPV-CUT |
| Node Classification | AIFB | Accuracy | 86.11 | RR-GCN-PPV |
| Node Classification | MUTAG | Accuracy | 79.41 | RR-GCN-PPV |
| Node Classification | DMGFULL | Accuracy | 63.38 | RR-GCN-PPV |
| Node Classification | DMGFULL | Accuracy | 57.52 | R-GCN |
| Node Classification | AM | Accuracy | 91.31 | RR-GCN-PPV-CUT (Unimportant relations removed) |
| Node Classification | AM | Accuracy | 84.8 | RR-GCN-PPV-CUT |
| Node Classification | AM | Accuracy | 84.65 | RR-GCN-PPV |
| Node Classification | MDGENRE | Accuracy | 67.33 | R-GCN |
| Node Classification | MDGENRE | Accuracy | 67.15 | RR-GCN-PPV |
| Node Classification | DMG777K | Accuracy | 63.97 | RR-GCN-PPV |
| Node Classification | DMG777K | Accuracy | 62.51 | R-GCN |
| Node Classification | BGS | Accuracy | 84.14 | RR-GCN-PPV-CUT |
| Node Classification | BGS | Accuracy | 78.97 | RR-GCN-PPV |