Semantic Textual Similarity on KorSTS (test)
77.43Spearman CorrelationKOMBO_Jamo
Evaluation Results
| Method | Links | |
|---|---|---|
| KOMBO_JamoTokenization=Jamo, Vocab Size=1702026.04 | 77.43 | |
| BERTTokenization=MorSubword, Vocab Size=32K2026.04 | 77.29 | |
| BERTTokenization=Subword, Vocab Size=32K2026.04 | 76.41 | |
| BERTTokenization=Character, Vocab Size=2K2026.04 | 76.34 | |
| KOMBO_StrokeTokenization=Stroke, Vocab Size=1302026.04 | 75.55 | |
| KOMBO_CjiTokenization=Cji, Vocab Size=1362026.04 | 75.28 | |
| BERTTokenization=Morpheme, Vocab Size=32K2026.04 | 74.35 | |
| BERTTokenization=Jamo, Vocab Size=1702026.04 | 72.96 | |
| KOMBO_BTSTokenization=BTS, Vocab Size=1122026.04 | 72.9 | |
| BERTTokenization=Stroke, Vocab Size=1302026.04 | 67.62 | |
| BERTTokenization=Word, Vocab Size=64K2026.04 | 65.5 | |
| BERTTokenization=Cji, Vocab Size=1362026.04 | 64.38 | |
| BERTTokenization=BTS, Vocab Size=1122026.04 | 60.14 |