distance_measure: <list[dot_product | cosine]> (Optional)
Description
Distance measure when measuring distance between node vector embeddings in link prediction. Using cosine similarity (cosine) instead of dot product (dot_product) tends to reduce popularity bias and tends to increase the diversity of predictions, often at the expense of accuracy.
Supported Task Types
- Link prediction
Default Values
| run_mode | Default Value |
|---|---|
| FAST | [dot_product] |
| NORMAL | [dot_product] |
| BEST | [dot_product] |