Semantic Textual Similarity on KorSTS (dev)
84.47Spearman Correlation (100x)KOMBO_Jamo
Evaluation Results
| Method | Links | |
|---|---|---|
| KOMBO_JamoTokenization=Jamo, Vocab Size=1702026.04 | 84.47 | |
| BERTTokenization=MorSubword, Vocab Size=32K2026.04 | 84.26 | |
| BERTTokenization=Character, Vocab Size=2K2026.04 | 83.98 | |
| BERTTokenization=Subword, Vocab Size=32K2026.04 | 83.8 | |
| KOMBO_CjiTokenization=Cji, Vocab Size=1362026.04 | 83.37 | |
| KOMBO_StrokeTokenization=Stroke, Vocab Size=1302026.04 | 82.42 | |
| BERTTokenization=Morpheme, Vocab Size=32K2026.04 | 82.19 | |
| KOMBO_BTSTokenization=BTS, Vocab Size=1122026.04 | 79.73 | |
| BERTTokenization=Jamo, Vocab Size=1702026.04 | 77.22 | |
| BERTTokenization=Word, Vocab Size=64K2026.04 | 72.2 | |
| BERTTokenization=Stroke, Vocab Size=1302026.04 | 67.99 | |
| BERTTokenization=Cji, Vocab Size=1362026.04 | 63.86 | |
| BERTTokenization=BTS, Vocab Size=1122026.04 | 56.63 |