Table 4 Ablation study of pretext tasks on node classification where the intra- and inter- are abbreviations of intra-metapath and inter-metapath contrasts, and local and global indicate node-node and node-graph contrasts, respectively.
From: A multi-view contrastive learning for heterogeneous network embedding
Intra- | Inter- | Local | Global | ACM | DBLP | IMDB | Aminer | FreeBase |
|---|---|---|---|---|---|---|---|---|
\(\surd\) | – | \(\surd\) | – | 89.21 | 91.94 | 58.71 | 84.95 | 69.61 |
\(\surd\) | – | – | \(\surd\) | 83.21 | 90.95 | 48.56 | 83.42 | 69.38 |
\(\surd\) | – | \(\surd\) | \(\surd\) | 90.52 | 92.52 | 60.12 | 86.17 | 71.16 |
– | \(\surd\) | \(\surd\) | \(\surd\) | 88.32 | 92.44 | 59.83 | 85.92 | 70.65 |
\(\surd\) | \(\surd\) | \(\surd\) | \(\surd\) | 91.02 | 93.29 | 60.75 | 86.63 | 71.41 |
