LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention

Benchmark Model Rank Results
common-sense-reasoning-on-recordLUKE 483M#5EM: 90.6F1: 91.2
entity-typing-on-open-entity-1MLMET#1F1: 78.2
named-entity-recognition-ner-on-conll-2003LUKE 483M#2F1: 94.3
named-entity-recognition-on-conllLUKE(Large)#2F1: 95.89
question-answering-on-squad11LUKE (single model)#1EM: 90.202F1: 95.379
question-answering-on-squad11LUKE#2EM: 90.2
question-answering-on-squad11LUKE 483M#47F1: 95.4
question-answering-on-squad11-devLUKE#2EM: 89.8
question-answering-on-squad11-devLUKE 483M#41F1: 95
question-answering-on-squad20LUKE (single model)#10EM: 87.429F1: 90.163
question-answering-on-squad20LUKE 483M#24F1: 90.2
relation-classification-on-tacred-1LUKE 483M#14F1: 72.7
relation-extraction-on-tacredLUKE#35F1 (5% Few-Shot): 51.6F1 (1% Few-Shot): 17.0