"The highlighted tokens are primarily Korean morphemes, words, or short phrases that serve as key semantic units, often marking grammatical functions, core vocabulary, or important content within sentences. These include nouns, verbs, adjectives, and particles, as well as compound words and set expressions. The activations tend to focus on tokens that carry the main informational or structural load in each phrase, such as subjects, objects, predicates, and descriptive elements."
Score Type | Accuracy | Precision | Recall | F1 score | TPR | TNR | FPR | FNR |
---|---|---|---|---|---|---|---|---|
detection | 0.6 | 1.0 | 0.2 | 0.333 | 0.2 | 1.0 | 0.0 | 0.8 |
fuzz | 0.65 | 0.895 | 0.34 | 0.493 | 0.34 | 0.96 | 0.04 | 0.66 |