CLEAR: Contrastive Learning for Sentence Representation

Benchmark Model Rank Results
linguistic-acceptability-on-colaMLM+ del-span+ reorder–Accuracy: 64.3%
natural-language-inference-on-qnliMLM+ subs+ del-span–Accuracy: 93.4%
natural-language-inference-on-rteMLM+ del-span–Accuracy: 79.8%
question-answering-on-quora-question-pairsMLM+ subs+ del-span–Accuracy: 90.3%
semantic-textual-similarity-on-mrpcMLM+ del-word+ reorder–Accuracy: 90.6%
semantic-textual-similarity-on-sts-benchmarkMLM+ del-word–Pearson Correlation: 0.905
sentiment-analysis-on-sst-2-binary-classificationMLM+ del-word+ reorder–Accuracy: 94.5