KELM: Knowledge Enhanced Pre-Trained Language Representations with Message Passing on Hierarchical Relational Graphs

Open paper
Benchmark Model Rank Results
common-sense-reasoning-on-recordKELM (finetuning RoBERTa-large based single model)#6EM: 89.1F1: 89.6
common-sense-reasoning-on-recordKELM (finetuning BERT-large based single model)#10EM: 76.2F1: 76.7
question-answering-on-copaKELM (finetuning BERT-large based single model)#35Accuracy: 78.0
question-answering-on-multircKELM (finetuning BERT-large based single model)#13F1: 70.8EM: 27.2

Rank counts only results with a code link.