Why do the authors use long text in the link prediction task, whereas only node text is used in the others?

yao8839836 / kg-bert

KG-BERT: BERT for Knowledge Graph Completion

Apache License 2.0

679 stars 141 forks source link

Why do the authors use long text in the link prediction task, whereas only node text is used in the others? #42

Open sa5r opened 8 months ago

sa5r commented 8 months ago

In the FB15K, the long node description is used for the link prediction task, whereas only the given node short text is used in the other tasks. Does this affect the consistency of the evaluation?