"The highlighted tokens are primarily Chinese characters with high semantic content, often representing nouns, adjectives, or verbs central to the meaning of the sentence, such as objects, actions, or descriptive qualities. There is also frequent activation on corrupted or unknown tokens, likely due to encoding errors or missing data, which are treated as significant by the model."
Score Type | Accuracy | Precision | Recall | F1 score | TPR | TNR | FPR | FNR |
---|---|---|---|---|---|---|---|---|
detection | 0.68 | 0.846 | 0.44 | 0.579 | 0.44 | 0.92 | 0.08 | 0.56 |
fuzz | 0.79 | 0.746 | 0.88 | 0.807 | 0.88 | 0.7 | 0.3 | 0.12 |