Word Similarity on SimLex-999
55.2Spearman CorrelationOurs (+post-process)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ours (+post-process)Type=Context-to-Vec, Post-processing=With post-processing2022.10 | 55.2 | — | |
| Ours (preliminary)Type=Context-to-Vec, Post-processing=None2022.10 | 54.9 | — | |
| BERT+Skip-gramType=Context-to-Vec2022.10 | 49.4 | — | |
| XLNetType=Contextualized, Representation level=word2022.10 | 49 | — | |
| BERTType=Contextualized, Representation level=token2022.10 | 48.9 | — | |
| GPT2Type=Contextualized, Representation level=token2022.10 | 48.4 | — | |
| XLNetType=Contextualized, Representation level=token2022.10 | 48.1 | — | |
| XLNetType=Contextualized, Representation level=avg2022.10 | 47.3 | — | |
| GPT2Type=Contextualized, Representation level=avg2022.10 | 46.9 | — | |
| BERTType=Contextualized, Representation level=avg2022.10 | 46.8 | — | |
| SynGCNType=Context-to-Vec2022.10 | 45.5 | — | |
| ROOTROOT-CAp=0, k (dimensionality)=3000, Training Corpus (Text8, BNC, Wiki052024)=Wiki0520242024.05 | 44.5 | — | |
| ELMoType=Contextualized, Representation level=avg2022.10 | 43.6 | — | |
| ELMoType=Contextualized, Representation level=token2022.10 | 41.7 | — | |
| ELMoType=Contextualized, Representation level=word2022.10 | 40.6 | — | |
| DepsType=Static2022.10 | 39.6 | — | |
| ContextLSTMType=Context-to-Vec2022.10 | 39.3 | — | |
| FASTTEXTType=Static2022.10 | 38.2 | — | |
| CBOWType=Static2022.10 | 38 | — | |
| Skip-gramType=Static2022.10 | 34.9 | — | |
| ROOTROOT-CAp=0, k (dimensionality)=700, Training Corpus (Text8, BNC, Wiki052024)=Text82024.05 | 32.1 | — | |
| Skip-gram(context)Type=Static2022.10 | 32 | — | |
| GloVeType=Static2022.10 | 31.6 | — | |
| GloVe+GTEmbedding Method=GloVe, Transform=Gaussian Transform, Refinement Corpus=text82018.10 | 27 | — | |
| GPT2Type=Contextualized, Representation level=word2022.10 | 26.4 | — | |
| GloVeEmbedding Method=GloVe2018.10 | 26 | — | |
| W2GEmbedding Method=W2G, Training Corpus=ukWaC and WaCkypedia2018.10 | 25 | — | |
| EllEmbedding Method=Ell, Training Corpus=ukWaC and WaCkypedia2018.10 | 24 | — | |
| W2V*text8Embedding Method=W2V, Training Corpus=text82018.10 | 22 | — | |
| BERTType=Contextualized, Representation level=word2022.10 | 13.4 | — | |
| GloVe*text8Embedding Method=GloVe, Training Corpus=text82018.10 | 13 | — | |
| ROOTROOT-CAp=0, Training Corpus (Text8, BNC, Wiki052024)=Total2024.05 | — | 1.176 |